Talks by visitors to the Department
2023 talks
- Robust Autonomous Vehicle Localization using GPS: from Tandem Drifting Cars to "GPS" on the Moon
Abstract:Speaker: Prof. Grace Gao
Date: 2023-12-05 Time: 12:00:00 (IST) Venue: SIT #001 Autonomous vehicles often operate in complex environments with various sensing uncertainties. On Earth, GPS signals can be blocked or reflected by buildings; and camera measurements are susceptible to lighting conditions. While having a variety of sensors is beneficial, including more sensing information can introduce more sensing failures as well as more computational load. For space applications, such as localization on the Moon, it can be even more challenging. In this talk, I will present our recent research efforts on robust vehicle localization under sensing uncertainties. We turn sensing noise and even absence of sensing into useful navigational signals. Inspired by cognitive attention in humans, we select a subset of "attention landmarks" from sensing measurements to reduce computation load and provide robust positioning. I will also show our localization techniques that enable various applications, from autonomous tandem drifting cars to a GPS-like system for the Moon.
Bio:Grace X. Gao is an assistant professor in the Department of Aeronautics and Astronautics at Stanford University. She leads the Navigation and Autonomous Vehicles Laboratory (NAV Lab). Prof. Gao has won a number of awards, including the National Science Foundation CAREER Award, the Institute of Navigation Early Achievement Award and the RTCA William E. Jackson Award. Prof. Gao and her students won Best Presentation of the Session/Best Paper Awards 29 times at Institute of Navigation conferences over the past 17 years. She also won various teaching and advising awards, including the Illinois College of Engineering Everitt Award for Teaching Excellence, the Engineering Council Award for Excellence in Advising, AIAA Illinois Chapter's Teacher of the Year, and most recently Advisor of the Year Award and Teacher of the Year Award by AIAA Stanford Chapter in 2022 and 2023, respectively.
- Robust Autonomous Vehicle Localization using GPS: from Tandem Drifting Cars to "GPS" on the Moon
Abstract:Speaker: Prof. Grace Gao
Date: 2023-12-05 Time: 12:00:00 (IST) Venue: SIT- 001 Autonomous vehicles often operate in complex environments with various sensing uncertainties. On Earth, GPS signals can be blocked or reflected by buildings; and camera measurements are susceptible to lighting conditions. While having a variety of sensors is beneficial, including more sensing information can introduce more sensing failures as well as more computational load. For space applications, such as localization on the Moon, it can be even more challenging. In this talk, I will present our recent research efforts on robust vehicle localization under sensing uncertainties. We turn sensing noise and even absence of sensing into useful navigational signals. Inspired by cognitive attention in humans, we select a subset of "attention landmarks" from sensing measurements to reduce computation load and provide robust positioning. I will also show our localization techniques that enable various applications, from autonomous tandem drifting cars to a GPS-like system for the Moon.
Bio:Grace X. Gao is an assistant professor in the Department of Aeronautics and Astronautics at Stanford University. She leads the Navigation and Autonomous Vehicles Laboratory (NAV Lab). Prof. Gao has won a number of awards, including the National Science Foundation CAREER Award, the Institute of Navigation Early Achievement Award and the RTCA William E. Jackson Award. Prof. Gao and her students won Best Presentation of the Session/Best Paper Awards 29 times at Institute of Navigation conferences over the past 17 years. She also won various teaching and advising awards, including the Illinois College of Engineering Everitt Award for Teaching Excellence, the Engineering Council Award for Excellence in Advising, AIAA Illinois Chapter's Teacher of the Year, and most recently Advisor of the Year Award and Teacher of the Year Award by AIAA Stanford Chapter in 2022 and 2023, respectively.
- Formal Methods for Software Reliability and Synthesis
Abstract:Speaker: Ashish Mishra
Date: 2023-11-23 Time: 11:30:00 (IST) Venue: MS Teams Building reliable software has been a classical goal in Computer Science. The most basic premise of my research is derived from this goal; Can we make programs safe and reliable using formal techniques while making programming as a discipline more democratic and accessible to the masses?
In this talk, I will begin by highlighting some of these overarching research interests and directions. I will primarily present two of my recent works highlighting the effective use of Refinement types, DSLs, and SMT-based techniques for the verification and synthesis of programs.
(i) The first is a new specification-guided synthesis procedure that uses Hoare-style pre- and post-conditions to express fine-grained effects of potential library component candidates to drive a bi-directional synthesis search strategy. It integrates a conflict-driven learning procedure into the synthesis algorithm that provides a semantic characterization of previously encountered unsuccessful search paths used to prune possible candidates' space as synthesis proceeds.
(ii) The second work is a new Refinement-Type system called Coverage Type which adapts the recent work in Incorrectness Logic to the specification and automated verification of test input generators used in modern property-based testing systems. Specifications are expressed in the language of refinement types, augmented with coverage types, types that reflect underapproximate constraints on program behavior.
Bio:Ashish Mishra is a Postdoctoral Researcher at Purdue University, where he works with Suresh Jagannathan in the areas of Programming Languages, Program Verification, and Program Synthesis. Ashish obtained his Ph.D. from the Indian Institute of Science, where he worked under the supervision of Y. N. Srikant. In addition to his work in Computer Science, Ashish is also interested in applying technology to public policies and solving social problems. He is currently involved with several Indian NGOs such as PARI (People's Archive for Rural India), Mosali (a startup trying to bring women into workforce), and others that are involved in Media Monitoring and Research.
- Stochastic Window Mean-payoff Games
Abstract:Speaker: Shibashis Guha, TIFR Bombay
Date: 2023-11-22 Time: 12:00:00 (IST) Venue: #501, Bharti Building Stochastic two-player games model systems with an environment that is both adversarial and stochastic. The environment is modeled by a player (Player 2) who tries to prevent the system (Player 1) from achieving its objective. We consider finitary versions of the traditional mean-payoff objective, replacing the long-run average of the payoffs by payoff average computed over a finite sliding window. Two variants have been considered: in one variant, the maximum window length is fixed and given, while in the other, it is not fixed but is required to be bounded. For both variants, we present complexity bounds and algorithmic solutions for computing strategies for Player 1 to ensure that the objective is satisfied with positive probability, with probability 1, or with probability at least p, regardless of the strategy of Player 2. The solution crucially relies on a reduction to the special case of non-stochastic two-player games. We give a general characterization of prefix-independent objectives for which this reduction holds. The memory requirement for both players in stochastic games is also the same as in non-stochastic games by our reduction. Moreover, for non-stochastic games, we improve upon the upper bound for the memory requirement of Player 1 and upon the lower bound for the memory requirement of Player 2.
This is a joint work with Laurent Doyen and Pranshu Gaba.
Bio:Shibashis Guha (https://www.tifr.res.in/~shibashis.guha/)
- Reasoning with interactive guidance
Abstract:Speaker: Dr.Niket Tandon
Date: 2023-11-21 Time: 11:00:00 (IST) Venue: SIT- 001 Humans view AI as a tool that listens and learns from their interactions, but this differs from the standard train-test paradigm. The goal of this talk is to introduce a step towards bridging this gap by enabling large language models to focus on human needs and continuously learn. Drawing inspiration from the theory of recursive reminding in Psychology, we propose a memory architecture to guide models to avoid repeating past errors. The talk discusses four essential research questions: who to ask, what to ask, when to ask and how to apply the obtained guidance. In tasks such as moral reasoning, planning, and other reasoning as well as benchmark tasks, our approach enables models to improve with reflection.
Bio:Niket Tandon is a Senior Research Scientist at the Allen Institute for AI in Seattle. His research interests are in commonsense reasoning and natural language guided reasoning. He works at the Aristo team responsible for creating AI which aced science exams. He obtained his Ph.D. from the Max Planck Institute for Informatics in Germany in 2016, where he was supervised by Professor Gerhard Weikum, resulting in the largest automatically extracted commonsense knowledge base at the time, called WebChild. He is also the founder of PQRS research, which introduces undergraduate students from underrepresented institutes to AI research. More information from him is available here: https://niket.tandon.info/
- Online List Labeling: Leveraging Predictions for Data Structures
Abstract:Speaker: Dr. Shikha Singh
Date: 2023-11-06 Time: 04:00:00 (IST) Venue: Bharti 501 A growing line of work shows how learned predictions can be used to break through worst-cast barriers to improve the running time of an algorithm. However, incorporating predictions into data structures with strong theoretical guarantees remains underdeveloped. This talk describes recent results on how predictions can be leveraged in the fundamental online list labeling problem. In the problem, n items arrive over time and must be stored in sorted order in an array of size Θ(n). The array slot of an element is its label and the goal is to maintain sorted order while minimizing the total number of elements moved (i.e., relabeled).
Bio:Shikha Singh is an Assistant Professor of Computer Science at Williams College. She obtained her PhD in Computer Science from Stony Brook University in 2018 and her Integrated MSc. in Mathematics and Computing from Indian Institute of Technology Kharagpur in 2013. Shikha's research is in the area of algorithms, with a focus on cache-efficient and scalable data structures, as well as algorithmic game theory.
- Training with Talk: A Machine Learning Makeover
Abstract:Speaker: Shashank Srivastava
Date: 2023-10-19 Time: 12:00:00 (IST) Venue: SIT- 001 Language and learning are deeply intertwined in humans. For example, in schools, we rely on processes such as reading books, listening to lectures, and engaging in student-teacher dialogs. In this talk, we will explore some recent work on building automated learning systems that can learn new tasks through natural language interactions with their users. We will cover multiple scenarios in this general direction: learning classifiers from language-based supervision, learning web-based tasks from explained demonstrations; and investigating when pretraining on language imparts effective inductive biases to large language models.
Bio:Shashank Srivastava is an assistant professor in the Computer Science department at the University of North Carolina (UNC) Chapel Hill. Shashank received his PhD from the Machine Learning department at CMU in 2018, and was an AI Resident at Microsoft Research in 2018-19. Shashank's research interests lie in conversational AI, interactive machine learning and grounded language understanding. Shashank has an undergraduate degree in Computer Science from IIT Kanpur, and a Master's degree in Language Technologies from CMU. He received the Yahoo InMind Fellowship for 2016-17. His research has been covered by popular media outlets including GeekWire and New Scientist.
- Socially aware Natural Language Processing
Abstract:Speaker: Snigdha Chaturvedi
Date: 2023-10-18 Time: 12:00:00 (IST) Venue: SIT- 001 NLP systems have made tremendous progress in recent years but lack a human-like language understanding. This is because there is a deep connection between language and people-- most text is created by people and for people. Despite this strong connection between language and people, existing NLP systems remain incognizant of the social aspects of language. In this talk, I describe three ways of designing Socially aware NLP systems. In the first part of the talk, I describe our socially aware approach to story generation by incorporating social relationships between various people to be mentioned in the story. We use a latent-variable-based approach that generates the story by conditioning on relationships to be exhibited in the story text using the latent variable. This latent variable-based design results in a better and explainable generation process. In the second part of the talk, I briefly describe our work on uncovering inherent social bias in automatically generated stories. We use a commonsense engine to reveal how such stories learn and amplify implicit social biases, especially gender biases. In the last part of the talk, I discuss methods to alleviate social biases. Specifically, I discuss debiasing text representations grounded in information theory. Using the rate-distortion function we show how we can remove information about sensitive attributes like race or gender from pre-trained text representations. This approach can successfully remove undesirable information while being robust to non-linear probing attacks.
Bio:Snigdha Chaturvedi is an Assistant Professor of Computer Science at the University of North Carolina, Chapel Hill. She specializes in Natural Language Processing, emphasizing narrative-like and socially aware understanding, summarization, and generation of language. Previously, she was an Assistant Professor at UC-Santa Cruz, and a postdoctoral fellow at UIUC and UPenn working with Dan Roth. She earned her Ph.D. in Computer Science from UMD in 2016, where she was advised by Hal Daume III. Her research has been supported by NSF, Amazon, and IBM.
- Securing Processors against Side-Channel Attacks: CPU Caches, Schedulers, and Beyond!
Abstract:Speaker: Prof. Gururaj Saileshwar
Date: 2023-10-13 Time: 11:00:00 (IST) Venue: SIT- 001 In recent years, micro-architectural side-channel attacks have emerged
as a unique and potent threat to security and privacy. Identifying these
side-channels is difficult as they often originate from undocumented
hardware structures, which are hidden from the software. Moreover, their
root-cause lies in crucial hardware performance optimizations, making
low overhead mitigation challenging. This talk will focus on both
discovery of new attacks and new low-cost defenses.
Bio:Gururaj Saileshwar is an Assistant Professor at the University of
Toronto, Dept of Computer Science. His research bridges computer
architecture and systems security, with interests including
micro-architectural side-channels, DRAM Rowhammer attacks, and trusted
execution environments. His past work has received an IEEE HPCA Best
Paper Award, an IEEE Micro Top Picks Honorable Mention, and his PhD
dissertation has been recognized with an IEEE HOST Best PhD Dissertation
Award and an IEEE TCCA / ACM SIGARCH Best Dissertation Award Honorable
Mention. His work appears in computer architecture conferences like
ASPLOS, MICRO, HPCA, and ISCA, and security conferences like USENIX
Security, IEEE S&P and CCS
- Computational modeling of neuronal dynamics
Abstract:Speaker: Parul Verma
Date: 2023-10-06 Time: 09:30:00 (IST) Venue: Online (MS Teams Link) Understanding neuronal dynamics, and how they are affected in neurological disorders, is one of the key problems in neuroscience today. This talk will describe advances in theoretical and biophysically grounded tools to understand neuronal mechanisms, with a focus on the functional activity of the entire brain. Specifically, it will demonstrate a graph-based mathematical model that captures the spectral and spatial features of the brain’s functional activity. This modeling approach revealed biophysical alterations in Alzheimer’s disease, different stages of sleep, and spontaneous fluctuations in electrophysiological functional activity. Together, these results aim to highlight the importance of such modeling techniques in identifying the underlying biophysical mechanisms of neuronal dynamics, which can be intractable to infer using neuroimaging data alone.
Bio:Parul Verma is a postdoc at the University of California San Francisco, Department of Radiology, since 2020. She obtained her Ph.D. at Purdue University. Before that, she obtained her B.Tech in Chemical Engineering from IIT Bombay. Parul’s doctoral work has been recognized by a faculty lectureship award from Purdue Chemical Engineering, and her postdoctoral work has been awarded a fellowship by the Alzheimer’s association.
- Computational modeling of neuronal dynamics
Abstract:Speaker: Parul Verma
Date: 2023-10-06 Time: 09:30:00 (IST) Venue: MS Teams Understanding neuronal dynamics, and how they are affected in neurological disorders, is one of the key problems in neuroscience today. This talk will describe advances in theoretical and biophysically grounded tools to understand neuronal mechanisms, with a focus on the functional activity of the entire brain. Specifically, it will demonstrate a graph-based mathematical model that captures the spectral and spatial features of the brain’s functional activity. This modeling approach revealed biophysical alterations in Alzheimer’s disease, different stages of sleep, and spontaneous fluctuations in electrophysiological functional activity. Together, these results aim to highlight the importance of such modeling techniques in identifying the underlying biophysical mechanisms of neuronal dynamics, which can be intractable to infer using neuroimaging data alone.
Bio:Parul Verma is a postdoc at the University of California San Francisco, Department of Radiology, since 2020. She obtained her Ph.D. at Purdue University. Before that, she obtained her B.Tech in Chemical Engineering from IIT Bombay. Parul’s doctoral work has been recognized by a faculty lectureship award from Purdue Chemical Engineering, and her postdoctoral work has been awarded a fellowship by the Alzheimer’s association.
- Deep Sensing: Jointly Optimizing Imaging and Processing
Abstract:Speaker: Dr. Sudhakar Kumawat,
Date: 2023-10-03 Time: 11:00:00 (IST) Venue: #001, SIT Building In this seminar, I will talk about the area of deep sensing where we jointly optimize the imaging (camera) parameters along with the deep learning models for novel computer vision applications. I will begin by discussing our recently published work "Action Recognition From a Single-Coded Image" where I will present our proposed framework for recognizing human actions directly from coded exposure images, without reconstructing the original scene. Next, I will talk about deep sensing in a broader context, discussing the motivation behind pursuing this research area, key ideas, and its application to existing and novel vision applications. Finally, I will briefly discuss how we are using deep sensing for a novel computer vision application called "Multimodal Material Segmentation in Road Scene Images".
Bio:Sudhakar Kumawat is a post-doctoral fellow at the Institute of Datability, Osaka University, Japan. He received his PhD from IIT Gandhinagar under the supervision of Dr. Shanmuganathan Raman. He was a TCS research fellow during PhD. Before that, he received his Integrated Dual Degree (B.Tech+M.Tech, 5 years) from the Computer Science and Engineering Department, IIT (BHU) Varanasi, in 2014. His broad area of research is computer vision, with a special interest in privacy-preserving computer vision, compressive sensing, and domain generalization. He has published papers in top computer vision journals and conferences such as TPAMI, ECCV, CVPR, and ICASSP. He received the best paper runner-up award at NCVPRIPG 2019.
- The Story of AWS Glue (VLDB 2023)
Abstract:Speaker: Mohit Saxena
https://www.amazon.science/publications/the-story-of-aws-glue
Date: 2023-09-27 Time: 12:00:00 (IST) Venue: SIT-001 AWS Glue is Amazon's serverless data integration cloud service that makes it simple and cost effective to extract, clean, enrich, load, and organize data. Originally launched in August 2017, AWS Glue began as an extract-transform-load (ETL) service designed to relieve developers and data engineers of the undifferentiated heavy lifting needed to load databases, data warehouses, and build data lakes on Amazon S3. Since then, it has evolved to serve a larger audience including ETL specialists and data scientists, and includes a broader suite of data integration capabilities. Today, hundreds of thousands of customers use AWS Glue every month.
Bio:Mohit Saxena is Senior Manager at Amazon Web Services. He leads the team that manages the serverless data integration service at Amazon globally. Earlier, he worked at IBM Research - Almaden and focused on database and storage systems. He completed his Ph.D in Computer Sciences from University of Wisconsin-Madison, M.S. from Purdue University and B.Tech in Computer Sciences from Indian Institute of Technology-Delhi.
- Bayesian spatiotemporal regression approaches for modelling and understanding the drivers of childhood vaccination outcomes
Abstract:Speaker: Sumeet Agarwal
Online (MS Teams): https://teams.microsoft.com/l/meetup-join/19%3a859e0622905d4a7980e595706e31fa0d%40thread.tacv2/1694307532808?context=%7b%22Tid%22%3a%22624d5c4b-45c5-4122-8cd0-44f0f84e945d%22%2c%22Oid%22%3a%22d147ea6a-9288-4db2-9f47-d243d61e426a%22%7d
Date: 2023-09-12 Time: 12:00:00 (IST) Venue: #001, SIT Building Incomplete immunisation coverage causes preventable illness and death in both developing and developed countries. Identification of factors that might modulate coverage could inform effective immunisation programmes and policies. We construct performance indicators to quantitatively approximate measures of the susceptibility of immunisation programmes to coverage losses, with an aim to identify correlations between trends in vaccine coverage and socioeconomic factors. We undertook a data-driven time-series analysis to examine trends in coverage of diphtheria, tetanus, and pertussis (DTP) vaccination across 190 countries over the past 30 years. We grouped countries into six world regions and used Gaussian process regression to forecast future coverage rates and provide a vaccine performance index: a summary measure of the strength of immunisation coverage in a country. Our vaccine performance index highlighted countries at risk of failing to achieve the global target of 90% coverage by 2015, and could aid policy makers' assessments of the strength and resilience of immunisation programmes. Subsequently, we have undertaken more localised analyses of vaccination coverage and confidence for India, including the development of novel latent-variable Bayesian hierarchical approaches for the inference of unobserved behavioural and social drivers of vaccination; we also discuss some outcomes from this ongoing work.
Bio:Sumeet Agarwal teaches in the areas of Electrical Engineering, Artificial Intelligence, and Cognitive Science at IIT Delhi. His research interests are focused around the use of machine learning and statistical modelling techniques to better understand the structure, function, and evolution of complex systems, in both the biological and the social sciences.
- Architectural Insights for Robustness and Fairness in Machine Learning
Abstract:Speaker: Professor Upamanyu Madhow
Date: 2023-09-05 Time: 12:00:00 (IST) Venue: Bharti 501 As data-driven machine-learnt algorithms become the technology of choice in an increasing array of applications, the research community recognizes the urgency of addressing shortcomings such as the lack of robustness (e.g., against adversarial examples and distribution shifts) and fairness (e.g., caused by bias in the training data). In this talk, we present two architectural insights, each based on a shift of perspective from the state of the art.
1) Software Architecture: We view the standard end-to-end paradigm for training DNNs, which does not provide explicit control over the features extracted by intermediate layers, as a fundamental bottleneck in the design of robust, interpretable DNNs. Motivated by ideas from communication theory (processing with matched filters) and neuroscience (neuronal competition), we propose adapting the training and inference framework for DNNs to provide more direct control over the shape of activations in intermediate layers. Preliminary results for the CiFAR-10 image database indicate significant gains in general-purpose robustness against noise and common corruptions, as well as against adversarial perturbations. We hope these results motivate further theoretical and experimental investigations: variants of the ideas we propose apply, in principle, to any DNN architecture or training model (supervised, unsupervised, self-supervised, semi-supervised).
2) Social Architecture: We view unfairness in DNNs resulting from data bias as a symptom of the unfairness and bias in the society from which the data is extracted. In an approach that is complementary to existing research on enhancing fairness during training and inference, we propose a framework for sequential decision-making aimed at dynamically influencing long-term societal fairness via positive feedback. We illustrate our ideas via a problem of selecting applicants from a pool consisting of two groups, one of which is under-represented, and hope that our results stimulate the collaboration between policymakers, social scientists and machine learning researchers required for real-world impact.
Bio:Upamanyu Madhow is Distinguished Professor of Electrical and Computer Engineering at the University of California, Santa Barbara. His current research interests focus on next generation communication, sensing and inference infrastructures, with emphasis on millimeter wave systems, and on fundamentals and applications of robust machine learning. Dr. Madhow is a recipient of the 1996 NSF CAREER award, co-recipient of the 2012 IEEE Marconi prize paper award in wireless communications, and recipient of a 2018 Distinguished Alumni award from the ECE Department at the University of Illinois, Urbana-Champaign. He is the author of two textbooks published by Cambridge University Press, Fundamentals of Digital Communication (2008) and Introduction to Communication Systems (2014). Prof. Madhow is co-inventor on 32 US patents, and has been closely involved in technology transfer of his research through several start-up companies, including ShadowMaps, a software-only approach to GPS location improvement which was deployed worldwide by Uber.
- Architectural Insights for Robustness and Fairness in Machine Learning
Abstract:Speaker: Professor Upamanyu Madhow , Department of Electrical & Computer Engineering, University of California, Santa Barbara
Date: 2023-09-05 Time: 12:00:00 (IST) Venue: #501, Bharti Building As data-driven machine-learnt algorithms become the technology of choice in an increasing array of applications, the research community recognizes the urgency of addressing shortcomings such as the lack of robustness (e.g., against adversarial examples and distribution shifts) and fairness (e.g., caused by bias in the training data). In this talk, we present two architectural insights, each based on a shift of perspective from the state of the art.
1) Software Architecture: We view the standard end-to-end paradigm for training DNNs, which does not provide explicit control over the features extracted by intermediate layers, as a fundamental bottleneck in the design of robust, interpretable DNNs. Motivated by ideas from communication theory (processing with matched filters) and neuroscience (neuronal competition), we propose adapting the training and inference framework for DNNs to provide more direct control over the shape of activations in intermediate layers. Preliminary results for the CiFAR-10 image database indicate significant gains in general-purpose robustness against noise and common corruptions, as well as against adversarial perturbations. We hope these results motivate further theoretical and experimental investigations: variants of the ideas we propose apply, in principle, to any DNN architecture or training model (supervised, unsupervised, self-supervised, semi-supervised).
2) Social Architecture: We view unfairness in DNNs resulting from data bias as a symptom of the unfairness and bias in the society from which the data is extracted. In an approach that is complementary to existing research on enhancing fairness during training and inference, we propose a framework for sequential decision-making aimed at dynamically influencing long-term societal fairness via positive feedback. We illustrate our ideas via a problem of selecting applicants from a pool consisting of two groups, one of which is under-represented, and hope that our results stimulate the collaboration between policymakers, social scientists and machine learning researchers required for real-world impact.
Bio:Upamanyu Madhow is Distinguished Professor of Electrical and Computer Engineering at the University of California, Santa Barbara. His current research interests focus on next generation communication, sensing and inference infrastructures, with emphasis on millimeter wave systems, and on fundamentals and applications of robust machine learning. Dr. Madhow is a recipient of the 1996 NSF CAREER award, co-recipient of the 2012 IEEE Marconi prize paper award in wireless communications, and recipient of a 2018 Distinguished Alumni award from the ECE Department at the University of Illinois, Urbana-Champaign. He is the author of two textbooks published by Cambridge University Press, Fundamentals of Digital Communication (2008) and Introduction to Communication Systems (2014). Prof. Madhow is co-inventor on 32 US patents, and has been closely involved in technology transfer of his research through several start-up companies, including ShadowMaps, a software-only approach to GPS location improvement which was deployed worldwide by Uber.
- Automated Decision Making for Safety Critical Applications
Abstract:Speaker: Prof. Mykel Kochenderfer (Stanford University)
Date: 2023-09-04 Time: 16:00:00 (IST) Venue: #501, Bharti Building Building robust decision making systems for autonomous systems is challenging. Decisions must be made based on imperfect information about the environment and with uncertainty about how the environment will evolve. In addition, these systems must carefully balance safety with other considerations, such as operational efficiency. Typically, the space of edge cases is vast, placing a large burden on human designers to anticipate problem scenarios and develop ways to resolve them. This talk discusses major challenges associated with ensuring computational tractability and establishing trust that our systems will behave correctly when deployed in the real world. We will outline some methodologies for addressing these challenges and point to some research applications that can serve as inspiration for building safer systems.
Bio:Mykel Kochenderfer is an Associate Professor of Aeronautics and Astronautics at Stanford University. He is the director of the Stanford Intelligent Systems Laboratory (SISL), conducting research on advanced algorithms and analytical methods for the design of robust decision making systems. Of particular interest are systems for air traffic control, unmanned aircraft, and automated driving where decisions must be made in uncertain, dynamic environments while maintaining safety and efficiency. Research at SISL focuses on efficient computational methods for deriving optimal decision strategies from high-dimensional, probabilistic problem representations. Prior to joining the faculty in 2013, he was at MIT Lincoln Laboratory where he worked on aircraft collision avoidance, leading to the creation of the ACAS X international standard for manned and unmanned aircraft. Prof. Kochenderfer is a co-director of the Center for AI Safety. He is an associate editor of the Journal of Artificial Intelligence Research and the Journal of Aerospace Information Systems. He is an author of the textbooks Decision Making under Uncertainty: Theory and Application (MIT Press, 2015), Algorithms for Optimization (MIT Press, 2019), and Algorithms for Decision Making (MIT Press, 2022).
- Deep Sensing: Jointly Optimizing Imaging and Processing
Abstract:Speaker: Dr. Sudhakar Kumawat
Date: 2023-09-03 Time: 11:00:00 (IST) Venue: SIT- 001 In this seminar, I will talk about the area of deep sensing where we jointly optimize the imaging (camera) parameters along with the deep learning models for novel computer vision applications. I will begin by discussing our recently published work "Action Recognition From a Single-Coded Image" where I will present our proposed framework for recognizing human actions directly from coded exposure images, without reconstructing the original scene. Next, I will talk about deep sensing in a broader context, discussing the motivation behind pursuing this research area, key ideas, and its application to existing and novel vision applications. Finally, I will briefly discuss how we are using deep sensing for a novel computer vision application called "Multimodal Material Segmentation in Road Scene Images".
Bio:Sudhakar Kumawat is a post-doctoral fellow at the Institute of Datability, Osaka University, Japan. He received his PhD from IIT Gandhinagar under the supervision of Dr. Shanmuganathan Raman. He was a TCS research fellow during PhD. Before that, he received his Integrated Dual Degree (B.Tech+M.Tech, 5 years) from the Computer Science and Engineering Department, IIT (BHU) Varanasi, in 2014. His broad area of research is computer vision, with a special interest in privacy-preserving computer vision, compressive sensing, and domain generalization. He has published papers in top computer vision journals and conferences such as TPAMI, ECCV, CVPR, and ICASSP. He received the best paper runner-up award at NCVPRIPG 2019.
- Gen-AI, its applications and end to end Development Life Cycle of AI solutions
Abstract:Speaker: Anupam Purwar
Date: 2023-09-01 Time: 16:00:00 (IST) Venue: Bharti 501 Generative AI in the natural language space is showing tremendous potential in automating various routine jobs. Recent studies have also demonstrated that Gen AI can aid with creative content creations. At the centre of this innovation in Gen AI are Large Language Models (LLMs), the leading ones are GPT 4, Claude2 and Llama 2 etc. Many of these LLMs are commercial, but there are open source ones too which can help organizations unlock tremendous value and help innovate. Through this talk, I would provide a practical way to develop an end to end application using LLMs in a scalable and affordable way. Speaker would cover software development life cycle for Generative AI solutions along with problem statement definition to help budding AI engineers, AI researchers and product managers alike.
Bio:Anupam Purwar is a Senior Research Scientist at Amazon Development Centre India, Hyderabad. He is a leading development of data science and machine learning-based solutions for Amazon Global Fulfillment. Anupam holds an MBA in Finance and Information Systems from the Indian School of Business and a Bachelor of Engineering from Birla Institute of Technology and Science, Pilani. He received merit scholarship and graduated among Top 5% in class both at ISB and BITS-Pilani. Prior to this, Anupam worked as a Research Scientist at Indian Institute of Science (IISc). At IISc, he was part of a multi-institutional effort which included IITs and ISRO to develop novel structures. Besides, he has authored 20+ peer reviewed articles pertaining to Machine Learning, IoT, computational design with 200+ citations and received multiple best paper awards. He is a certified Machine learning professional with 8+ certifications from Google and AWS.
- Fusing AI and Formal Methods for Automated Synthesis
Abstract:Speaker: Priyanka Golia
Date: 2023-08-28 Time: 12:00:00 (IST) Venue: MS Teams We entrust large parts of our daily lives to computer systems, which are becoming increasingly more complex. Developing scalable yet trustworthy techniques for designing and verifying such systems is an important problem. In this talk, our focus will be on automated synthesis, a technique that uses formal specifications to automatically generate systems (such as functions, programs, or circuits) that provably satisfy the requirements of the specification. I will introduce a state-of-the-art functional synthesis algorithm that leverages artificial intelligence to provide an initial guess for the system and then uses formal methods to repair and verify the guess to synthesize a system that is correct by construction. I will conclude by exploring the potential for combining AI and formal methods to address real-world scenarios.
Bio:Priyanka Golia has completed her Ph.D. in the joint degree program of NUS, Singapore and IIT Kanpur, India. Her research interests lie at the intersection of formal methods and artificial intelligence. In particular, her dissertation work has focused on designing scalable automated synthesis and testing techniques.
Her work has been awarded Best Paper Nomination at ICCAD-21 and Best Paper Candidate at DATE-23. She was named one of the EECS Rising Stars in 2022. She has co-presented a tutorial on Automated Synthesis: Towards the Holy Grail of AI at AAAI-22 and IJCAI-22, and she is co-authoring an upcoming book (on invitation from NOW publishers) on functional synthesis. - "Fast Multivariate Multipoint Evaluation over Finite Fields"
Abstract:Speaker: Dr. Sumanta Ghosh (Caltech) Online Talk
https://teams.microsoft.com/l/meetup-join/19%3ac00a05b5843f4486843ed7ca9c863eeb%40thread.tacv2/1692427565207?context=%7b%22Tid%22%3a%22624d5c4b-45c5-4122-8cd0-44f0f84e945d%22%2c%22Oid%22%3a%22870c4f19-6710-453d-a17d-f35f49b733e1%22%7d
Date: 2023-08-24 Time: 09:30:00 (IST) Multivariate multipoint evaluation is the problem of evaluating a multivariate polynomial, given as a coefficient vector, simultaneously at multiple evaluation points. The straightforward algorithm for this problem is to iteratively evaluate the input polynomial at each input point. The question of obtaining faster-than-naive (ideally, linear time) algorithms for this problem is a natural and fundamental question in computational algebra. Besides, fast algorithms for this problem are closely related to fast algorithms for other natural algebraic questions like polynomial factorization and modular composition.
Nearly linear time algorithms have been known for the univariate instance of multipoint evaluation for close to five decades due to the work of Borodin and Moenck. However, fast algorithms for the multivariate version have been much harder to come by. In a significant improvement to the state of art for this problem, Umans in 2008 and Kedlaya-Umans in 2011 gave nearly linear time algorithms for this problem over field of small characteristic and over all finite fields respectively, provided that the number of variables m is at most d^{o(1)} where the degree of the input polynomial in every variable is less than d. They also stated the question of designing fast algorithms for the large variable case as an open problem.
In this talk, we present two new algorithms for this problem. The first one is a nearly linear time (algebraic) algorithm for not-too-large fields of small characteristics. For the large variable case, this is the first nearly linear time algorithm for this problem over any large enough field. The second gives a nearly linear time (non-algebraic) algorithm over all finite fields.
The talk is based on joint work with Vishwas Bhargava, Zeyu Guo, Mrinal Kumar, Chandra Kanta Mohapatra, and Chris Umans.
Bio:Dr. Sumanta Ghosh (Caltech)
- Text image analysis from low resource dataset
Abstract:Speaker: Prof. Partha PratimRoy, IIT Roorkee
Date: 2023-08-23 Time: 12:00:00 (IST) Venue: #501, Bharti building Text image understanding has long been an active research area because of its complexity and challenges due to a variety of shapes. For bench-marking such a system, the dataset is a necessary and important resource to develop. The deep learning-based text image analytics tasks, such as detection and recognition have shown impressive results under the setting of full supervision of completely labeled datasets. However, the creation of such datasets with a large volume of samples is a challenging and time-consuming task. This research presentation will highlight a few solutions towards effective analysis of textual image analysis in scarcity of data annotation.
It is observed that the performance of the generic scene text detection method drops significantly due to the partial annotation of training data which introduces unnecessary noise. We propose a text region refinement method that provides robustness against the partially annotated training data in scene text detection. This approach works as a two-tier scheme. In the first tier text-probable regions apart from ground-truth text are obtained by applying hybrid loss. Next, these text-probable regions generate pseudo-labels to refine annotated text regions in the second tier during training. The proposed method exhibits a significant improvement over the baseline and existing approaches for the partially annotated training data.
Besides, recognition of textual images is a difficult task sometimes as sufficient labeling of data is not available for some unexplored scripts, especially Indic scripts. The design of deep neural network models makes it necessary to extend training datasets in order to introduce unseen variations. We propose an Adversarial Feature DeformationModule (AFDM) that learns ways to elastically warp extracted features in a scalable manner. The AFDM is inserted between intermediate layers and trained alternatively with the original framework, boosting its capability to better learn highly informative features rather than trivial ones. We record results for varying sizes of training data and observe that our enhanced network generalizes much better in the low-data regime.
Bio:Dr. Partha Pratim Roy (FIETE, SMIEEE) is presently working as an Associate Professor in the Department of Computer Science and Engineering, Indian Institute of Technology (IIT), Roorkee. He received his Masters in 2006 and Ph.D. in 2010 from Universitat Autonoma de Barcelona, Spain. He did postdoctoral stays in France and Canada from 2010 to 2013. Dr. Roy gathered industrial experience while working for about 4 years in TCS and Samsung. In Samsung, he was a part-leader of the Computer Vision research team. He is the recipient of the "Best Student Paper" awarded by the International Conference on Document Analysis and Recognition (ICDAR), 2009, Spain. He has published more than 200 research papers in various international journals, and conference proceedings. He has co-organized several international conferences and workshops, has been a member of the Program Committee of a number of international conferences, and acts as a reviewer for many journals in the field. His research interests include Pattern Recognition, Document Image Processing, Biometrics, and Human-Computer Interaction. He is presently serving as Associate Editor of ACM Transactions on Asian and Low-Resource Language Information Processing, Neurocomputing, IET Image Processing, IET Biometrics, IEICE Transactions on Information and Systems, and Springer Nature Computer Science.
- Principled Reinforcement Learning to Model our Dynamic Environments
Abstract:Speaker: Prof Chandrajit Bajaj
Date: 2023-08-11 Time: 12:00:00 (IST) Venue: #501, Bharti Building Can computers be programmed to learn to model progressive approximations of the underlying dynamical processes of specific environments, through interaction (i.e. spatio-temporal sensing) . We answer this in the affirmative with a proviso that not all the Hamiltonian models of environmental processes are learnable at optimal fidelity. Computers equipped with stable numerical solvers (some, possibly simultaneously learnable), are at the mercy of the noise and uncertainty of the sensed environmental observations. can nevertheless be programmed to stably train, cross-validate and test stochastic PDE (partial differential equation) neural operators. The learning is along optimally controlled pathways that satisfy a form of the Hamilton-Jacobi-Bellman equation. In this talk, I shall explain a framework of learning Hamiltonian models (Hamiltonians) as a partially observable controlled Markov decision process model (COMDP) and based on the Pontryagin's maximum principle. The COMDP model learning trajectory operates on a constrained manifold that satisfy the conservation laws of the underlying physics, via application of Noether's theorem. The COMDP includes learning dynamic stabilizing control satisfying learned Lyapunov functions for error bounded, convergent solutions, and additionally produces sparse approximations that avoid overfitting This talk shall show a few examples of such learned spatio-temporal models of dynamic environments, with various approximations of dynamic shape and function .
This is joint work with my students Taemin Heo, Minh Nguyen, Yi Wan
Bio:Chandrajit Bajaj, Department of Computer Science and Oden Institute, Center for Computational Visualization,University of Texas at Austin
http://www.cs.utexas.edu/~bajaj
bajaj@cs.utexas.edu, bajaj@oden.utexas.edu
- A Quantum Revolution in Computation
Abstract:Speaker: Umesh Vazirani (UC Berkeley)
Date: 2023-07-31 Time: 03:30:00 (IST) Venue: #501, Bharti Building We are well into the NISQ era of Noisy Intermediate Scale Quantum Computers. Four years on from Google's ‘quantum supremacy’ experiment, we have a deeper understanding of the nature of that experiment, the computing power of NISQ and novel techniques for benchmarking such computers and characterizing their error models. I will also describe how concepts from cryptography have provided novel and counter-intuitive ways of probing quantum systems and the prospects they hold for the next generation of quantum computers taking on the quantum supremacy challenge.
Bio:Prof Umesh V. Vazirani is the Roger A. Strauch Professor of Electrical Engineering and Computer Science at the University of California, Berkeley, and the director of the Berkeley Quantum Computation Center. His research interests lie primarily in quantum computing. Vazirani is one of the founders of the field of quantum computing. His 1993 paper with his student Ethan Bernstein on quantum complexity theory defined a model of quantum Turing machines which was amenable to complexity based analysis. This paper also gave an algorithm for the quantum Fourier transform, which was then used by Peter Shor within a year in his celebrated quantum algorithm for factoring integers. With Charles Bennett, Ethan Bernstein, and Gilles Brassard, he showed that quantum computers cannot solve black-box search problems faster than O({sqrt {N}}) in the number of elements to be searched. This result shows that the Grover search algorithm is optimal. It also shows that quantum computers cannot solve NP-complete problems in polynomial time using only the certifier. He is also a co-author of a textbook on algorithms. Vazirani was awarded the Fulkerson Prize for 2012 for his work on improving the approximation ratio for graph separators and related problems (jointly with Satish Rao and Sanjeev Arora). In 2018, he was elected to the National Academy of Sciences.
- Hardness of Testing Equivalence to Sparse Polynomials Under Shifts
Abstract:Speaker: Suryajith Chillara
Date: 2023-07-26 Time: 03:00:00 (IST) Venue: #001, SIT Building We say that two given polynomials $f, g in R[x_1, ldots, x_n]$, over a ring $R$, are equivalent under shifts if there exists a vector $(a_1, ldots, a_n)in R^N$ such that $f(x_1+a_1, ldots, x_n+a_n) = g(x_1, ldots, x_n)$. This is a special variant of the polynomial projection problem in Algebraic Complexity Theory. That is, instead of being given two polynomials $f$ and $g$ as input as described before, we are just given a polynomial $f$ and a parameter $t$ and we are interested in studying a more general problem of testing equivalence of $f$ to any of the polynomials in $R[x_1, ldots, x_n]$ which have at most $t$ many monomials with non-zero coefficients.
Grigoriev and Karpinski (FOCS 1990), Lakshman and Saunders (SIAM J. Computing, 1995), and Grigoriev and Lakshman (ISSAC 1995) studied the problem of testing polynomial equivalence of a given polynomial to any $t$-sparse polynomial, over the rational numbers, and gave exponential time algorithms. In the past two decades, these exponential time algorithms could not be improved and this is a major motivation behind our study of hardness of this problem.
We show that $SparseShift_R$ is at least as hard as checking if a given system of polynomial equations over $R[x_1,ldots, x_n]$ has a solution (Hilbert's Nullstellensatz). We also study the gap versions of this problem and show NP-hardness for certain regime of parameters.
Our results to some extent throws a light on why this problem in general has been evading the efforts to provide efficient algorithms.
Joint work with Coral Grichener (Google, Israel) and Amir Shpilka (Tel Aviv University).
Link: https://drops.dagstuhl.de/opus/volltexte/2023/17674/
Bio:Suryajith Chillara (https://suryajith.github.io/) IIIT Hyderabad
- Some results in the Intersection of Game Theory and Logic
Abstract:Speaker: Ramit Das
Date: 2023-06-27 Time: 16:00:00 (IST) Venue: Bharti Bldg. #404 + Team Links We shall address the issues of modelling or formalising game theoretic properties like Nash Equilibrium, Finite Improvement Property, Weak Acyclicity of various game forms in various kinds of logic. We shall investigate the expressive powers offered by each logic, the model checking theorems and also a completeness proof of a decidable logic variant. We hope that this investigation would have an impact on the formalisation of game theory and its allied areas like computational social choice theory.
Bio:Ramit Das is a researcher in Theoretical Computer Science about to get his PhD from the Institute of Mathematical Sciences, Chennai. He is interested in understanding the nature of computation in its various forms. For his PhD, he trained in Mathematical Logic and applied it to aspects of strategic games. His academic interests lie in trying to build bridges in Game Theory, Logic and areas of Complexity Theory like Descriptive Complexity Theory.
- Approximate Model Counting: Is SAT Oracle More Powerful than NP Oracle?
Abstract:Speaker: Gunjan Kumar
Date: 2023-06-15 Time: 11:00:00 (IST) Venue: Bharti Bldg. #404 + Team Links Given a Boolean formula $phi$ over $n$ variables, the problem of model counting is to compute the number of solutions of $phi$. Model counting is a fundamental problem in computer science with wide-ranging applications in domains such as quantified information leakage, probabilistic reasoning, network reliability, neural network verification, and more. Owing to the #P-hardness of the problems, Stockmeyer initiated the study of the complexity of approximate counting and showed that $log n$ calls to an NP oracle are necessary and sufficient to achieve tight guarantees.
It is well known that an NP oracle does not fully capture the behavior of SAT solvers, as SAT solvers are also designed to provide satisfying assignments when a formula is satisfiable, without additional overhead. Accordingly, the notion of SAT oracle has been proposed to capture the behavior of SAT solver wherein given a Boolean formula, an SAT oracle returns a satisfying assignment if the formula is satisfiable or returns unsatisfiable otherwise.
The primary contribution of this work is to study the relative power of the NP oracle and SAT oracle in the context of approximate model counting. We develop a new methodology to achieve the main result: a SAT oracle is no more powerful than an NP oracle in the context of approximate model counting.
(To appear in ICALP 2023. Joint work with Diptarka Chakraborty, Sourav Chakraborty, and Kuldeep S Meel).
Bio:Gunjan Kumar did his B.Tech in Computer Science and Engineering from the Indian Institute of Technology, Guwahati. Thereafter, he pursued his MS and Ph.D. from Tata Institute of Fundamental Research, Mumbai, and currently, he is a postdoctoral researcher at the National University of Singapore. His broad area of interest is Algorithms and Complexity with a focus on sublinear algorithms.
- Measuring and Improving the Internal Conceptual Representations of Deep Learning
Abstract:Speaker: Ramakrishna Vedantam
Online (MS Teams): https://teams.microsoft.com/l/meetup-join/19%3ac00a05b5843f4486843ed7ca9c863eeb%40thread.tacv2/1685675828429?context=%7b%22Tid%22%3a%22624d5c4b-45c5-4122-8cd0-44f0f84e945d%22%2c%22Oid%22%3a%22d147ea6a-9288-4db2-9f47-d243d61e426a%22%7d
Date: 2023-06-08 Time: 12:00:00 (IST) Venue: #501, Bharti Building Endowing machines with abstract, flexible conceptual representations and the ability to combine known concepts to make novel, "conceptual-leaps" is a long-standing goal of artificial intelligence (AI). In pursuit of this goal, I will discuss my works on the foundations of concept learning for deep learning models. In particular I will focus on: multimodal learning (to ground concept representations more precisely into the world), quantifying robustness (to assess if atomic concepts are learnt correctly) and machine reasoning (to combine known atomic concepts into novel, emergent ones). Finally, I will speculate on important research directions to pursue for realizing the promise of general, robust and human interpretable AI systems.
Bio:Ramakrishna Vedantam received the Ph.D. Degree in Computer Science from the Georgia Institute of Technology in 2018, before joining Facebook AI Research (FAIR) in New York. Currently, he is a visiting researcher at the New York University (NYU) center for data science (CDS). Rama's current research interests are around the foundations of robustness, multimodal learning and reasoning with large-scale deep learning models. At Georgia Tech, Rama's Ph.D. research was supported by the highly competitive Google Ph.D. fellowship. During his Ph.D. Rama also spent time at Google Research in Mountain View, Facebook AI Research in Menlo Park, Microsoft Research in Cambridge, UK and Ecole Centrale in Paris working on various topics at the intersection of probabilistic deep learning, multimodal learning, and reasoning. Rama developed the CIDEr metric popularly used in the AI community for evaluating vision and language models, and has published his research at various top-tier AI/ML venues such as ICML, NeurIPS, ICLR, CVPR, ICCV and EMNLP.
- Semi-nonparametric Demand Estimation in the Presence of Unobserved Factors
Abstract:Speaker: Ashwin Venkataraman
Date: 2023-05-29 Time: 15:10:00 (IST) Venue: SIT-001 Discrete choice models are commonly used to model customer demand because of their ability to capture substitution patterns in customer choices.
Demand predictions from these models are then used as inputs in key operational decisions for firms such as what collection of products to show to customers or what prices to charge for different products in order to maximize revenues. In many applications of discrete choice modeling, there exist unobserved factors (UFs) driving the consumer demand that are not included in the model. Ignoring such UFs when fitting the choice model can produce biased parameter estimates, leading to poor demand predictions and suboptimal decisions. At the same time, accounting for UFs during estimation is challenging since we typically have only partial or indirect information about them. Existing approaches such as the classical BLP estimator (Berry et al. 1995) make strong parametric assumptions to deal with this challenge, and therefore can suffer from model misspecification issues when the assumptions are not met in practice.In this talk, I'll present a novel semi-nonparametric estimator for dealing with UFs in the widely used mixture of logit choice model that does not impose any parametric assumptions on the mixing distribution or the underlying mechanism generating the UFs. We theoretically characterize the benefit of using our estimator over the BLP estimator, and leverage the alternating minimization framework to design an efficient algorithm that implements our proposed estimator. Using a simulation study, we demonstrate that our estimator is robust to different ground-truth settings, whereas the performance of the BLP estimator suffers significantly under model misspecification. Finally, using real-world grocery sales data, we show that accounting for product and store-level UFs can significantly improve the accuracy of predicting weekly demand at an individual product and store level, with an avg. 57% improvement across 12 product categories over a state-of-the-art benchmark that ignores UFs during estimation.
Joint work with Prof. Srikanth Jagabathula and Sandeep Chitla, both from NYU Stern School of Business.
Bio:Ashwin Venkataraman is an assistant professor of operations management at the Naveen Jindal School of Management at the University of Texas at Dallas (UTD). His research interests lie at the intersection of machine learning, operations management, and marketing, with a focus on developing novel models and methodologies that can leverage the vast amounts of customer data that firms have access to nowadays. Prior to joining UTD, Ashwin received an MS and PhD in computer science from the Courant Institute of Mathematical Sciences at New York University, and his doctoral thesis won an honorable mention (joint-second place) in the 2019 INFORMS Dantzig Dissertation Award. Before joining graduate school, Ashwin completed a B.Tech in Computer Science and Engineering from IIT Delhi.
- The Road Not Taken: Exploring Alias Analysis Based Optimizations Missed by the Compiler
Abstract:Speaker: Dr. Piyush Kedia
Date: 2023-04-21 Time: 03:00:00 (IST) Venue: #113, SIT Building The alias analysis aims to answer whether two pointers can overlap during runtime. However, static alias analysis is imprecise. Because the alias analysis is used by many compiler optimizations, including loop transformations, the program's performance may suffer, especially in the presence of loops, due to the imprecision of alias analysis.
In this talk, I'll present our tool, Scout, which can disambiguate two-pointers at runtime using single memory access. The key idea is to constrain the allocation size and alignment during memory allocations to enable fast disambiguation checks. Our technique enabled new opportunities for loop-invariant code motion, dead store elimination, loop vectorization, and load elimination in an already optimized code. Our performance improvements are up to 51.11% for Polybench benchmarks and up to 0.89% for SPEC benchmarks.
Bio:Piyus Kedia is an assistant professor at IIIT Delhi. He received his Ph.D. from IIT Delhi. He works in the area of programming languages and systems security.
- On The Membership Problem for Hypergeometric Sequences with Rational Parameters
Abstract:Speaker: Klara Nosan.
Date: 2023-04-19 Time: 03:00:00 (IST) Venue: #001, SIT Building We investigate the Membership Problem for hypergeometric sequences: given a hypergeometric sequence ⟨u_n⟩ of rational numbers and a rational value t, decide whether t occurs in the sequence. We show decidability of this problem under the assumption that in the defining recurrence f(n) u_{n+1} = g(n) u_n, the roots of the polynomials f and g are all rational numbers. We further show the problem remains decidable if the splitting fields of the polynomials f and g are distinct or if f and g are monic polynomials that both split over a quadratic number field.
Our proof relies on bounds on the density of primes in arithmetic progressions. We also observe a relationship between the decidability of the Membership problem (and variants) and the Rohrlich-Lang conjecture in transcendence theory.
This talk is based on works done in collaboration with George Kenison, Amaury Pouly, Mahsa Shirmohammadi and James Worrell.
Bio:https://www.irif.fr/~nosan/
- Backdoor Attacks in Computer Vision: Challenges in Building Trustworthy Machine Learning Systems
Abstract:Speaker: Dr. Aniruddha Saha
Date: 2023-04-19 Time: 12:00:00 (IST) Venue: #001, SIT Building Deep Neural Networks (DNNs) have become the standard building block in numerous machine learning applications. The widespread success of these networks has driven their deployment in sensitive domains like health care, finance, autonomous driving, and defense-related applications.
However, DNNs are vulnerable to adversarial attacks. Research has shown that an adversary can tamper with the training process of a model by injecting misrepresentative data (poisons) into the training set. The manipulation is done in a way that the victim's model will malfunction only when a trigger modifies a test input. These are called backdoor attacks. For instance, a backdoored model in a self-driving car might work accurately for days before it suddenly fails to detect a pedestrian when the adversary decides to exploit the backdoor.
In this talk, I will show ways in which state-of-the-art deep learning methods for computer vision are vulnerable to backdoor attacks and a few defense methods to remedy the vulnerabilities. Optimizing only for accuracy is not enough when we are developing machine learning systems for high stakes domains. Making machine learning systems trustworthy is our biggest challenge in the next few years.
Bio:Aniruddha Saha is currently a Postdoctoral Associate with the Center for Machine Learning (CML) in the University of Maryland Institute for Advanced Computer Studies (UMIACS). He received his PhD in Computer Science from the University of Maryland, Baltimore County. His research interests include Computer Vision, Adversarial Robustness, Data Poisoning, Backdoor Attacks and Trustworthy Machine Learnin
- Towards effective human-robot collaboration in shared autonomy systems
Abstract:Speaker: Raunak Bhattacharyya
Date: 2023-03-31 Time: 12:00:00 (IST) Venue: #501, Bharti Building Automated agents have the potential to augment human capabilities in safety-critical applications such as driving, service and inspection, and smart manufacturing. As the field of robotics and AI is quickly emerging, one critical and challenging problem is ensuring that autonomous agents can collaborate and interact with humans. In this talk, I will present our work on how automated agents can model human decision making, plan around human operators, and explain their decisions. First, I will present an approach based on imitation learning to model real-world human behavior and demonstrate its application to model human driving trajectories. Second, I will present a hybrid data-driven and rule-based approach to generate novel scenarios which can be used for planning. Third, I will present ongoing work on explainable automated agents. Finally, I will discuss my future research plan centered around shared autonomous systems which includes optimally allocating authority between automated and human controllers, learning from imperfect demonstrations, metacognition for human-robot collaboration, and safe autonomous planning in the presence of humans.
Bio:Dr. Raunak Bhattacharyya is a Postdoctoral Research Associate with the Oxford Robotics Institute, University of Oxford. His research focuses on human-autonomy interaction in shared autonomy systems. Raunak completed his Ph.D. at Stanford University, where he was a doctoral researcher in the Stanford Intelligent Systems Lab. He earned two Master's Degrees, in Computer Science and in Aerospace Engineering from Georgia Tech, and an undergraduate degree in Aerospace Engineering from IIT Bombay. Raunak received the Postdoctoral Enrichment Award from the Alan Turing Institute, UK, and the Graduate Research Award from the Transportation Research Board, USA.
Online Link (MS Teams): https://teams.microsoft.com/l/meetup-join/19%3a859e0622905d4a7980e595706e31fa0d%40thread.tacv2/1679902599494?context=%7b%22Tid%22%3a%22624d5c4b-45c5-4122-8cd0-44f0f84e945d%22%2c%22Oid%22%3a%22d147ea6a-9288-4db2-9f47-d243d61e426a%22%7d
- Security for the Internet of Things: Challenges and Prospects
Abstract:Speaker: Dr. Shantanu Pal, Assistant Professor, School of Information Technology, Deakin University, Melbourne, Australia
Date: 2023-03-29 Time: 12:00:00 (IST) Venue: Online (MS Teams Link) This talk aims to report security mechanisms for large-scale Internet of Things (IoT) systems, in particular, the need for access control, identity management, delegation of access rights and the provision of trust within such systems. The talk will discuss the design and development of an access control architecture for the IoT. How the policy-based approach provides a fine-grained access for authorized users to services while protecting valuable resources from unauthorized access will be discussed in detail. The talk will also explore an identity-less, asynchronous and decentralized delegation model for the IoT leveraging the advantage of blockchain technology. This further calls for better designing the IoT infrastructures, optimizing human engagement, managing IoT identity, and advocating lightweight access control solutions in the broad context of IoT to create a fertile ground for research and innovation. This talk will also discuss various challenges, including the propagation of uncertainty in IoT networks and prospects of IoT access control mechanisms using emerging technologies, e.g., blockchain.
Bio:Dr. Shantanu Pal is an Assistant Professor in the School of Information Technology, Deakin University, Melbourne, Australia. Shantanu holds a PhD in Computer Science from Macquarie University, Sydney, Australia. He was a Research Fellow at the Queensland University of Technology (QUT), Brisbane, Australia. He was also an associate researcher working with CSIRO's Data61, Australia. Shantanu's research interests are the Internet of Things (IoT), access control, blockchain technology, big data and distributed applications for cyber-physical systems, mobile and cloud computing, uncertainty propagation in IoT networks, emerging technologies, e.g., machine learning and artificial intelligence, etc. Shantanu is listed in the world's top 2% of scientists according to the recently released list by Stanford University, USA, in 2022 in Computer Networking and Communications.
- Repeatedly Matching Items to Agents Fairly and Efficiently
Abstract:Speaker: Shivika Narang (PhD student at IISc)
Date: 2023-01-27 Time: 16:00:00 (IST) Venue: bharti-501 We consider a novel setting where a set of items are matched to the same set of agents repeatedly over multiple rounds. Each agent gets exactly one item per round, which brings interesting challenges to finding efficient and/or fair repeated matchings. A particular feature of our model is that the value of an agent for an item in some round depends on the number of rounds in which the item has been used by the agent in the past. We present a set of positive and negative results about the efficiency and fairness of repeated matchings. For example, when items are goods, a variation of the well-studied fairness notion of envy-freeness up to one good (EF1) can be satisfied under certain conditions. Furthermore, it is intractable to achieve fairness and (approximate) efficiency simultaneously, even though they are achievable separately. For mixed items, which can be goods for some agents and chores for others, we propose and study a new notion of fairness that we call swap envy-freeness (swapEF).
Joint work with Prof Ioannis Caragiannis.
https://arxiv.org/abs/2207.01589
Bio:Shivika Narang is a PhD student, and recipient of the Tata Consultancy Services (TCS) Research Scholarship at the Indian Institute of Science, Bengaluru, where she is a member of the Game Theory Lab. She is being advised by Prof Y Narahari. She is broadly interested in Algorithmic Game Theory and Approximation Algorithms. Her current work is focused on fairness in matchings and allocations.
- Fusing AI and Formal Methods for Automated Synthesis.
Abstract:Speaker: Priyanka Golia (IITK & NUS)
Date: 2023-01-17 Time: 15:00:00 (IST) Venue: bharti-501 We entrust large parts of our daily lives to computer systems, which are becoming increasingly more complex. Developing scalable yet trustworthy techniques for designing and verifying such systems is an important problem. In this talk, our focus will be on automated synthesis, a technique that uses formal specifications to automatically generate systems (such as functions, programs, or circuits) that provably satisfy the requirements of the specification. I will introduce a state-of-the-art synthesis algorithm that leverages artificial intelligence to provide an initial guess for the system, and then uses formal methods to repair and verify the guess to synthesize probably correct system. I will conclude by exploring the potential for combining AI and formal methods to address real-world scenarios.
Bio:Priyanka Golia is a final year Ph.D. candidate at NUS, Singapore and IIT Kanpur. Her research interests lie at the intersection of formal methods and artificial intelligence. In particular, her dissertation work has focused on designing scalable automated synthesis and testing techniques. Her work has been awarded Best Paper Nomination at ICCAD-21 and Best Paper Candidate at DATE-23. She was named one of the EECS Rising Stars in 2022. She has co-presented a tutorial on Automated Synthesis: Towards the Holy Grail of AI at AAAI-22 and IJCAI-22, and She is co-authoring an upcoming book (on invitation from NOW publishers) on functional synthesis.
- Selection in the Presence of Biases
Abstract:Speaker: Prof. Nisheeth Vishnoi (Yale University)
Date: 2023-01-09 Time: 15:00:00 (IST) Venue: SIT-001 In selection processes such as hiring, promotion, and college
admissions, biases toward socially-salient attributes of candidates
are known to produce persistent inequality and reduce aggregate
utility for the decision-maker. Interventions such as the Rooney Rule
and its generalizations, which require the decision maker to select at
least a specified number of individuals from each affected group, have
been proposed to mitigate the adverse effects of such biases in
selection.
In this talk, I will discuss recent works which have established that
such lower-bound constraints can be effective in improving aggregate
utility.
Papers:
https://arxiv.org/abs/2001.08767
https://arxiv.org/abs/2010.10992
https://arxiv.org/abs/2202.01661
Bio:Nisheeth Vishnoi is a professor of computer science and a co-founder of the Computation and Society Initiative at Yale University. His research focuses on foundational problems in computer science, machine learning, and optimization. He is also broadly interested in understanding and addressing some of the key questions that arise in nature and society from a computational viewpoint. Here, his current focus is on physics-inspired algorithms and algorithmic fairness. He is the author of two monographs and the book Algorithms for Convex Optimization.
He was the recipient of the Best Paper Award at IEEE FOCS in 2005, the IBM Research Pat Goldberg Memorial Award in 2006, the Indian National Science Academy Young Scientist Award in 2011, the IIT Bombay Young Alumni Achievers Award in 2016, and the Best Technical Paper award at ACM FAT* in 2019. He was elected an ACM Fellow in 2019. - Towards Next-Generation ML/AI: Robustness, Optimization, Privacy.
Abstract:Speaker: Krishna Pillutla, visiting researcher (postdoc) at Google Research, USA
Date: 2023-01-06 Time: 12:00:00 (IST) Venue: bharti-501 Two trends have taken hold in machine learning and artificial intelligence: a move to massive, general-purpose, pre-trained models and a move to small, on-device models trained on distributed data. Both these disparate settings face some common challenges: a need for (a) robustness to deployment conditions that differ from training, (b) faster optimization, and (c) protection of data privacy.
As a result of the former trend, large language models have displayed emergent capabilities they have not been trained for. Recent models such as GPT-3 have attained the ability to generate remarkably human-like long-form text. I will describe Mauve, a measure to quantify the goodness of this emergent capability. It measures the gap between the distribution of generated text and that of human-written text. Experimentally, Mauve correlates the strongest with human evaluations of the generated text and can quantify a number of its qualitative properties.
The move to massively distributed on-device federated learning of models opens up new challenges due to the natural diversity of the underlying user data and the need to protect its privacy. I will discuss how to reframe the learning problem to make the model robust to natural distribution shifts arising from deployment on diverse users who do not conform to the population trends. I will describe a distributed optimization algorithm and show how to implement it with end-to-end differential privacy.
To conclude, I will discuss my ongoing efforts and future plans to work toward the next generation of ML/AI by combining the best of both worlds with applications to differentially private language models and text generation to decentralized learning.
Bio:Krishna Pillutla is a visiting researcher (postdoc) at Google Research, USA in the Federated Learning team. He obtained his Ph.D. at the University of Washington where he was advised by Zaid Harchaoui and Sham Kakade. Before that, he received his M.S. from Carnegie Mellon University and B.Tech. from IIT Bombay where he was advised by Nina Balcan and J. Saketha Nath respectively. Krishna's research has been recognized by a NeurIPS outstanding paper award (2021) and a JP Morgan Ph.D. fellowship (2019-20).
2022 talks
- The present developments and future challenges of near-eye displays for augmented and virtual reality
Abstract:Speaker: Dr. Praneeth Chakravarthula, Research Scholar at Princeton University
Date: 2022-11-25 Time: 11:00:00 (IST) Venue: Bharti-501 Holography and eye tracking have become the holy grail of display technology with their combined promise of offering unprecedented capabilities in near-eye displays. In this talk, I will discuss the developments in holography and eye tracking, and the challenges to overcome before such near-eye displays can be practical.
Bio:Praneeth is a research scholar at Princeton University working on end-to-end human and machine perception, with a special focus on computational cameras and displays. His research interests lie at the intersection of optics, perception, graphics, optimization, and machine learning. Praneeth obtained his Ph.D. from UNC Chapel Hill, under the advice of Prof. Henry Fuchs, on everyday-use eyeglasses style near-eye displays for virtual and augmented reality. His research has won several awards including recent best paper awards at SIGGRAPH 2022 and ISMAR 2022.
- On Creative Visual Communication
Abstract:Speaker: Prof. Karan Singh from the University of Toronto
Date: 2022-11-14 Time: 11:00:00 (IST) Venue: Bharti-501 This talk describes the fundamental human need for creative visual communication and discusses research on techniques that lie at the intersection of art, mathematics, interaction and AI. The talk will cover open challenges in the expression of shape and motion from sketch and audio-visual input, and its perception both in 2D and AR/VR, highlighting various interfaces such as cassie, crossshade+true2form, www.ilovesketch.com, www.meshmixer.com www.flatfab.com, colorsandbox.com, janusxr.org and jaliresearch.com that are creative solutions to problems in sketch and color modeling, AR/VR creation and facial animation.
Bio:Karan Singh is a Professor in Computer Science (since 2002) at the University of Toronto. He holds several CS degrees: a BTech. (1991) from the Indian Institute of Technology Madras and MS (1992), PhD (1995) from the Ohio State University. His research interests lie in interactive graphics, spanning geometric and anatomic modeling, visual perception, character and facial animation, sketch/touch based interfaces and interaction techniques for Augmented and Virtual Reality (AR/VR). He has been a development lead on the technical Oscar (2003) winning modeling and animation system Maya, and co-founded multiple companies including sketch2 (acquired by MRI Software 2021), JanusXR, and JALI. He supervised the design and research of critically acclaimed research systems ILoveSketch, MeshMixer (acquired by Autodesk in 2011), Neobarok and FlatFab. Karan co-directs a globally reputed graphics and HCI lab, DGP, has over 150 peer-reviewed publications, and has supervised over 50 MS/PhD/ postdoctoral students. He was the R&D Director for the 2004 Oscar winning animated short film Ryan and has had an exhibition of electronic art titled Labyrinths. His research on audio-driven facial animation JALI, was used to animate speech for all characters in the AAA game Cyberpunk 2077, and his recent research on anatomically animated faces Animatomy, will be showcased in the upcoming film Avatar: The Way of Water.
- Towards Autonomous Driving in Dense, Heterogeneous, and Unstructured Environments
Abstract:Speaker: Dr. Rohan Chandra,postdoctoral researcher at the University of Texas, Austin
Date: 2022-11-09 Time: 12:00:00 (IST) Venue: Bharti-501 In this talk, I discuss many key problems in autonomous driving towards handling dense, heterogeneous, and unstructured traffic environments. Autonomous vehicles (AV) at present are restricted to operating on smooth and well-marked roads, in sparse traffic, and among well-behaved drivers. I present new techniques to perceive, predict, and navigate among human drivers in traffic that is significantly denser in terms of number of traffic-agents, more heterogeneous in terms of size and dynamic constraints of traffic agents, and where many drivers may not follow the traffic rules and have varying behaviors. My talk is structured along three themes—perception, driver behavior modeling, and planning. More specifically, I will talk about:
1. Improved tracking and trajectory prediction algorithms for dense and heterogeneous traffic using a combination of computer vision and deep learning techniques.
2. A novel behavior modeling approach using graph theory for characterizing human drivers as aggressive or conservative from their trajectories.
3. Behavior-driven planning and navigation algorithms in mixed and unstructured traffic environments using game theory and risk-aware planning.
Finally, I will conclude by discussing the future implications and broader applications of these ideas in the context of social robotics where robots are deployed in warehouses, restaurants, hospitals, and inside homes to assist human beings.
Bio:Rohan Chandra is currently a postdoctoral researcher at the University of Texas, Austin, hosted by Dr. Joydeep Biswas. Rohan obtained his B.Tech from the Delhi Technological University, New Delhi in 2016 and completed his MS and PhD in 2018 and 2022 from the University of Maryland advised by Dr. Dinesh Manocha. His doctoral thesis focused on autonomous driving in dense, heterogeneous, and unstructured traffic environments. He is a UMD’20 Future Faculty Fellow, RSS’22 Pioneer, and a recipient of a UMD’20 summer research fellowship. He has published his work in top computer vision and robotics conferences (CVPR, ICRA, IROS) and has interned at NVIDIA in the autonomous driving team. He has served on the program committee of leading conferences in robotics, computer vision, artificial intelligence, and machine learning. He has given invited talks at academic seminars and workshops and has served on robotics panels alongside distinguished faculty. Outside of research, Rohan enjoys playing board games, teaching, and mentoring younger students. Webpage: http://rohanchandra30.github.io/.
- Blending PDEs with Machine Learning for “Mechanics”
Abstract:Speaker: Somdatta Goswami is an Assistant Professor of Research in the Division of Applied Mathematics at Brown University
Date: 2022-11-03 Time: 17:00:00 (IST) Venue: Teams A new paradigm in scientific research has been established with the integration of data-driven and physics-informed methodologies in the domain of deep learning, and it is certain to have an impact on all areas of science and engineering. This field, popularly termed scientific machine learning, relies on a known model, some (or no) high-fidelity data, and partially known constitutive relationships or closures, to be able to close the gap between the physical models and the observational data. Despite the fact that these strategies have been effective in many fields, they still face significant obstacles, such as the need for accurate and precise knowledge transmission in a data-restricted environment, and the investigation of data-driven methodologies in the century-old field of mechanics is still only in its infancy. The application of deep learning techniques within the context of functional and operator regression to resolve PDEs in mechanics will be the major focus of this talk. The approaches' extrapolation ability, accuracy, and computing efficiency in big and small data regimes, including transfer learning, would serve as indicators of their effectiveness.
Bio:Somdatta Goswami is an Assistant Professor of Research in the Division of Applied Mathematics at Brown University. Her research is focused on the development of efficient scientific machine-learning algorithms for high-dimensional physics-based systems in the fields of computational mechanics and biomechanics. She joined Brown University as a postdoctoral research associate under the supervision of Prof. George Karniadakis in January 2021. Prior to that, she completed her Ph.D. at Bauhaus University in Germany under Prof. Timon Rabczuk's guidance. During this time, she worked on developing techniques to overcome the limitations of conventional numerical methods for phase field-based modeling of fracture with isogeometric analysis and machine learning approaches.
- Model Counting meets Distinct Elements
Abstract:Speaker: Dr. Kuldeep Meel, NUS Presidential Young Professorship in NUS
Date: 2022-11-01 Time: 16:00:00 (IST) Venue: Bharti-501 Constraint satisfaction problems (CSP) and data stream models are two powerful abstractions to capture a wide variety of problems arising in different domains of computer science. Developments in the two communities have mostly occurred independently and with little interaction between them. In this work, we seek to investigate whether bridging the seeming communication gap between the two communities may pave the way to richer fundamental insights.
In this talk, I will describe how our investigations lead us to observe striking similarity in the core techniques employed in the algorithmic frameworks that have evolved separately for model counting and Distinct Elements computation. We design a simple recipe for translation of algorithms developed for Distinct Elements to that of model counting, resulting in new algorithms for model counting. We then observe that algorithms in the context of distributed streaming can be transformed to distributed algorithms for model counting. We next turn our attention to viewing streaming from the lens of counting and show that framing Distinct Elements estimation as a special case of #DNF counting allows us to obtain a general recipe for a rich class of streaming problems, which had been subjected to case-specific analysis in prior works.
(Joint work with A. Pavan, N. V. Vinodchandran, A. Bhattacharyya. The paper appeared at PODS 2021, was invited to ACM TODS as “Best of PODS 2021”, and received 2022 ACM SIGMOD Research Highlight Award and CACM Research Highlights).
Bio:Kuldeep Meel holds the NUS Presidential Young Professorship in the School of Computing at the National University of Singapore (NUS). His research interests lie at the intersection of Formal Methods and Artificial Intelligence. He is a recipient of the 2022 ACP Early Career Researcher Award, the 2019 NRF Fellowship for AI and was named AI's 10 to Watch by IEEE Intelligent Systems in 2020. His research program's recent recognitions include the 2022 CACM Research Highlight Award, 2022 ACM SIGMOD Research Highlight, IJCAI-22 Early Career Spotlight, 2021 Amazon Research Award, "Best of PODS-21" invite from ACM TODS, "Best Papers of CAV-20" invite from FMSD journal, IJCAI-19 Sister conferences best paper award track invitation. Before joining NUS in 2017, he received M.S. and Ph.D. from Rice University, co-advised by Supratik Chakraborty and Moshe Y. Vardi. His thesis work received the 2018 Ralph Budd Award for Best Ph.D. Thesis in Engineering and the 2014 Outstanding Masters Thesis Award from Vienna Center of Logic and Algorithms, IBM Ph.D. Fellowship, and Best Student Paper Award at CP 2015. He graduated with Bachelor of Technology (with honors) in Computer Science and Engineering from IIT Bombay.
- Designing for Local knowledge Towards Inclusive Development
Abstract:Speaker: Dr. Deepika Yadav
Date: 2022-10-26 Time: 12:00:00 (IST) Venue: Teams Limited opportunities to learn, share and develop shared understandings of health are key impediments to sustainable and equitable growth of society. Decentring the process of knowledge construction to include local, domestic, cultural, and shared knowledge to complement expert-, clinical-, and individual-oriented ways has the potential to progress holistically towards achieving sustainable development goals, particularly in the areas which have lagged such as related to women and child health, reproductive health, and gender inequities. Examining these problems from the Human-Computer Interaction field perspective, I will present my research that focuses on designing and developing digital technologies for community and intimate health. I will first discuss findings from my research on improving learning opportunities for ASHAs who are lay village women working as community health workers in under-served regions of India. Taking an Action Research approach, my research responds to the training challenges and constraints in which ASHAs operate in rural India by involving healthcare workers and NGOs in the design and production of a novel interaction and communication channel that was successful in supporting the training of ASHAs. I proposed a low-cost Interactive Voice Response-based mobile training tool and through different field studies demonstrated the feasibility and acceptability of alternative peer-to-peer learning methods mediated through mobile technology. This includes providing training to 500+ ASHAs in Haryana and Delhi regions. Through my studies, I document the potential of the proposed tool to provide training and address queries that are culturally rooted and otherwise overlooked which have impacts on child and maternal health nationwide. I will then discuss my current research on intimate care that explores interaction design mechanisms to support self-discovery-based knowledge development and address interpersonal tensions in using digital intimate technologies in domestic and workplace settings.
Bio:Deepika is a Digital Futures postdoctoral fellow at Stockholm University and KTH Royal University. Her research interests include human-computer interaction, information and communication and development, and computer-supported collaborative work. Her work involves designing and developing digital technologies for community health with a focus on women's health and intimate health challenges in low-resource and shared contexts. She obtained her Ph.D. from IIIT Delhi in 2021. She is also the winner of the Grand Challenge Explorations Award from the Bill & Melinda Gates Foundation which supported her Ph.D. research on community health workers in India.
- Functional Synthesis - An Ideal Meeting Ground for Formal Methods and Machine Learning
Abstract:Speaker: Dr. Kuldeep Meel, NUS
Date: 2022-10-20 Time: 16:00:00 (IST) Venue: Bharti-501 Don't we all dream of the perfect assistant whom we can just tell what to do and the assistant can figure out how to accomplish the tasks? Formally, given a specification F(X,Y) over the set of input variables X and output variables Y, we want the assistant, aka functional synthesis engine, to design a function G such that F(X,G(X)) is true. Functional synthesis has been studied for over 150 years, dating back Boole in 1850's and yet scalability remains a core challenge. Motivated by progress in machine learning, we design a new algorithmic framework Manthan, which views functional synthesis as a classification problem, relying on advances in constrained sampling for data generation, and advances in automated reasoning for a novel proof-guided refinement and provable verification. The significant performance improvements call for interesting future work at the intersection of machine learning, constrained sampling, and automated reasoning.
Relevant publications: CAV-20, IJCAI-21, and ICCAD-21 (Best Paper Award nomination)
(Based on joint work with Priyanka Golia, Friedrich Slivovsky, and Subhajit Roy)
Bio:Kuldeep Meel holds the NUS Presidential Young Professorship in the School of Computing at the National University of Singapore (NUS). His research interests lie at the intersection of Formal Methods and Artificial Intelligence. He is a recipient of the 2022 ACP Early Career Researcher Award, the 2019 NRF Fellowship for AI and was named AI's 10 to Watch by IEEE Intelligent Systems in 2020. His research program's recent recognitions include the 2022 CACM Research Highlight Award, 2022 ACM SIGMOD Research Highlight, IJCAI-22 Early Career Spotlight, 2021 Amazon Research Award, "Best of PODS-21" invite from ACM TODS, "Best Papers of CAV-20" invite from FMSD journal, IJCAI-19 Sister conferences best paper award track invitation. Before joining NUS in 2017, he received M.S. and Ph.D. from Rice University, co-advised by Supratik Chakraborty and Moshe Y. Vardi. His thesis work received the 2018 Ralph Budd Award for Best Ph.D. Thesis in Engineering and the 2014 Outstanding Masters Thesis Award from Vienna Center of Logic and Algorithms, IBM Ph.D. Fellowship, and Best Student Paper Award at CP 2015. He graduated with Bachelor of Technology (with honors) in Computer Science and Engineering from IIT Bombay.
- Specification-Guided Reinforcement Learning
Abstract:Speaker: Dr. Suguman Bansal, Georgia Institute of Technology
Date: 2022-10-18 Time: 11:00:00 (IST) Venue: Teams Reinforcement Learning (RL), especially when combined with neural networks (NN), has made remarkable strides in control synthesis in real-world domains, including challenging continuous (infinite-state) environments in robotics and game-playing. Yet, current RL approaches are poorly suited for control synthesis for long-horizon tasks for a number of reasons. First, typically control tasks in RL are specified in the form of rewards. Providing a suitable reward function for complex, long-horizon tasks can be daunting. Second, RL algorithms are inherently myopic, as they respond to immediate rewards, which may cause the algorithm to learn optimal policies for the short term but below-par policies in the long term. Lastly, the learned policies offer a poor degree of assurance: Neither are these policies interpretable and nor can they be verified against the desired task.
In this talk, I will discuss our work on RL from logical specifications. Here the task is expressed in the form of temporal specifications as opposed to rewards. While the use of temporal specifications resolves the first issue, I will discuss our trials and triumphs in our quest to design RL algorithms that scale to long-horizon tasks and offer theoretical guarantees.This talk is based on our NeurIPS 2021 paper and Invited Contribution to Henzinger-60.
Bio:Suguman Bansal is an incoming Assistant Professor in the School of Computing at Georgia Institute of Technology, starting in January 2023. Her research is focused on formal methods and their applications to artificial intelligence, programming languages, and machine learning. Previously, she was an NSF/CRA Computing Innovation Postdoctoral Fellow at the University of Pennsylvania, mentored by Prof. Rajeev Alur. She completed her Ph.D. at Rice University, advised by Prof. Moshe Y. Vardi. She is the recipient of the 2020 NSF CI Fellowship and has been named a 2021 MIT EECS Rising Star.
- Fractional Stable Matching and Computational Social Choice
Abstract:Speaker: Dr. Sanjukta Roy, Postdoctoral scholar in College of IST at Penn State
Date: 2022-10-12 Time: 12:00:00 (IST) Venue: Teams In Computational Social Choice, we are generally presented with problems where we deal with agents with preferences. A classical problem in this domain is the stable matching problem where the input is a bipartite graph G and every vertex (i.e., an agent) has a complete and strict preference list over the vertices from the other side. A stable matching is a mapping f: E(G) mapsto {0,1} that satisfies some stability criteria.
In the Fractional Stable Matching problem the range of f is [0,1] such that the sum of values assigned to the edges incident to a vertex is at most 1. We study fractional matching with cardinal preferences. We consider more general stability notions: the so-called ordinal stability and cardinal stability.
We investigate the computational complexity of finding an ordinally stable or cardinally stable fractional matching which either maximizes the social welfare (i.e., the overall utilities of the agents) or the number of fully matched agents (i.e., agents whose matching values sum up to one). In this talk we will focus on the definitions and discuss our algorithmic findings. In the end, I will also discuss some of my other research interests.
Bio:Dr. Sanjukta Roy, Postdoctoral scholar in College of IST at Penn State
- The Multiple Facets of Human Navigation on the Web
Abstract:Speaker: Dr. Akhil Arora, EPFL
Date: 2022-10-06 Time: 11:00:00 (IST) Venue: Teams Navigability, defined as the ability to effectively and efficiently maneuver from a source to a target node, is one of the most important functions of any network. Fueled by our curiosity and information seeking nature, we usually navigate a plethora of real-world networks including but not limited to the Web, online encyclopedic systems, news articles, and social networks. Consequently, the navigation patterns employed by humans provide tremendous insights into the way in which humans explore, browse, and interface with information on the Web, and thus, understanding and modeling human navigation behavior is an important task in the broad field of web data science and data mining.
In this talk, we will look into three different facets of human navigation of networks, namely: (1) enablers, (2) characteristics, and (3) applications. We will first discuss the ability of entity linkers to enrich the link structure of information networks, thereby acting as an enabler of network navigation. We will also review Eigenthemes, a language and domain agnostic unsupervised method for entity linking. Next, we will review the characteristics of human navigation behavior in Wikipedia, which is the largest online encyclopedic system. Specifically, we will discuss how and under what scenarios do real human navigation patterns differ from synthetic patterns generated using standardized null models. Lastly, we will discuss an important application, that of operationalizing link insertion. To this end, we will showcase the utility of navigation patterns as a strong signal in identifying positions for inserting entities in Wikipedia articles.
Bio:Akhil Arora is a PhD student advised by Prof. Robert West of the Data Science Lab (dlab) at EPFL and an external research collaborator of the Wikimedia Foundation. Prior to this, Akhil spent close to five years in the industry working with the research labs of Xerox and American Express as a Research Scientist. Even before that, he graduated from the Computer Science department of IIT Kanpur in June 2013, where he was advised by Prof. Arnab Bhattacharya. Akhil’s research interests include large-scale data management, graph mining, and machine learning. He is a recipient of the prestigious “EDIC Doctoral Fellowship” for the academic year 2018-19, and the Most Reproducible Paper Award at the ACM SIGMOD Conference, 2018. He has published his research in prestigious data management conferences, served as a program committee member, and co-organized workshops in these conferences. Additional details are available on his website: https://dlab.epfl.ch/people/aarora/.
- Research at Computer Vision Lab at IIT Madras
Abstract:Speaker: Prof. Anurag Mittal, Professor, CSE, IITM
Date: 2022-05-05 Time: 12:00:00 (IST) Venue: Teams In this talk, I will give an overview of the research activities at the Computer Vision lab, IIT Madras, with a focus on some of our exciting latest work in Video Understanding and Vision + language. In particular, I will talk about the importance of fusing Geometry and Statistical Inference for building robust Computer Vision systems and our work in the multi-camera domain where integration of informaion from multiple cameras is required. Then, I will talk about our work in robust Feature Extraction, Representation and Matching that utilizes orders rather than raw intensities, as orders between pixels are much more
robust to changes in the illumination, object color etc. and lead to much more robust systems. Next, I will talk about our work in shape representation and matching. Shape is the most defining characteristic of most objects and leads to the most accurate object representations. It is particularly useful in 3D object representation and matching. Finally, I will talk in somewhat more detail about two recent lines of work. In the first, I will talk about our new work on a better representation for Videos using an attention mechanism that attends to the common regions across a video using the concepts of co-segmentation. In the second line of work, I will talk about improving vision+language tasks such as image and video captioning and visual question answering by using better models, including those that mitigate database biases.
Bio:Anurag Mittal is currently a professor in the Computer Science and Engg. dept. at IIT Madras heading the Computer Vision Lab (2006-present). Prior to this, from 2002-2005, he was a Research Scientist at the Real-Time Vision and Modeling Department at Siemens Corporate Research, Princeton NJ. He completed his PhD in Computer Science from the University of Maryland, College Park in Dec 2002. Before that, he completed an MS from Cornell University in 2000 and a B.Tech. in Computer Science and Engineering from the Indian Institute of Technology, Delhi in 1997.
His research interests are in all areas of Computer Vision, and he has published extensively on many varied topics such as Surveillance and Security, Feature Detection and Extraction, Robust Matching techniques, Shape Representation and Matching, zero-shot learning, Video Representations, Vision+Language and Image and Video Super-resolution. He has 5000+ citations to his research publications.
He is an active member of the Computer Vision community, and is an Associate Editor of CVIU since 2013. He has also acted as an Area Chair for many major conferences in the area such as ICCV, CVPR, ECCV and ACCV.
- Causal Machine Learning: A path to out-of-distribution generalization and fairness
Abstract:Speaker: Dr. Amit Sharma (MSR India)
Date: 2022-04-27 Time: 12:00:00 (IST) Venue: Teams Current machine learning models face a number of challenges, including out-of-distribution generalization, fairness, privacy, robustness, and explainability. I will present causal machine learning as a new paradigm for learning models that aims to address all these challenges through a unifying formal framework. The key principle in causal machine learning is to express data-generating assumptions as a graph, which then helps to identify the correct regularization for different learning goals. Theoretically, we can show that a predictive model that uses only graph parents of the target variable is invariant to data distribution shifts, more robust to arbitrary interventions, and has better privacy guarantees than a standard ML model.
I will describe applications of this principle in 1) obtaining state-of-the-art accuracy in the domain generalization task where the test data comes from different distribution than train; 2) detecting and fixing unfairness of machine learning systems.
Bio:Amit Sharma is a Principal Researcher at Microsoft Research India. His work bridges causal inference techniques with machine learning, with the goal of making machine learning models generalize better, be explainable and avoid hidden biases. To this end, Amit has co-led the development of the open-source DoWhy library for causal inference and DiCE library for counterfactual explanations. The broader theme in his work is how machine learning can be used for better decision-making, especially in sensitive domains. In this direction, Amit collaborates with NIMHANS on mental health technology, including a recent app, MindNotes, that encourages people to break stigma and reach out to professionals. His work has received many awards including a Best Paper Award at ACM CHI 2021 conference, Best Paper Honorable Mention at ACM CSCW 2016 conference, 2012 Yahoo! Key Scientific Challenges Award and the 2009 Honda Young Engineer and Scientist Award. Amit received his Ph.D. in computer science from Cornell University and B.Tech. in Computer Science and Engineering from Indian Institute of Technology (IIT) Kharagpur.
https://www.microsoft.com/en-us/research/people/amshar/
- Data-Driven Ecosystem Migration: Migrating R from Lazy to Strict Semantics
Abstract:Speaker: Dr. Aviral Goel Ph.D., CS, Northeastern University
Date: 2022-04-21 Time: 12:00:00 (IST) Venue: SIT-001 Evolving mainstream language ecosystems is challenging because of large packagerepositories and millions of active users. Even a tiny change can break a big
chunk of otherwise functional code at this scale, significantly impacting users
and discouraging adoption. Partial adoption of changes leads to incompatible
libraries causing fragmentation of the ecosystem. In this talk, I will propose a data-driven
strategy to evolve a language at scale with minimal impact on its users. I willapply this strategy to evolve the R language ecosystem from lazy to strict semantics.
Bio:Aviral Goel is a Computer Science Ph.D. candidate at Northeastern University. He
has a B.Tech. in Electronics and Communication Engineering from NSUT, New Delhi.
Before pursuing a Ph.D., he wrote software at Yahoo! and National Centre for
Biological Sciences, Bengaluru.He maintains a personal website at: http://aviral.io. - Modelling functional activity for brain state identification
Abstract:Speaker: Dr. Sukrit Gupta, Hasso Plattner Institute
Date: 2022-03-10 Time: 11:00:00 (IST) Venue: Bharti-501 Advances in neuroimaging techniques have made it possible to access intricate details of brain function and structure. Functional magnetic resonance imaging (MRI) gives us access to the brain's functional activations and aids in understanding its functional architecture during rest, neurological diseases, and varied task states. Data from functional MRI scans can be modelled as a network, known as the brain functional connectome, such that the network nodes represent brain regions and the edges between these nodes represent functional relationships between the regions. Studying the functional connectome has led to an improved understanding of cognitive and diseased brain states. In this presentation, I will discuss techniques that we proposed to uncover the functional alterations in the brain during different brain states.
Bio:Dr. Sukrit Gupta is currently working as a research fellow with Hasso Plattner Institute, Berlin. Previously, he was working as a Research Scientist with the Deep Learning for Medical Imaging division with the Institute of Infocomm Research, Agency for Science Technology and Research (A*STAR), Singapore. He obtained his PhD in Computer Science from Nanyang Technological University, Singapore in 2020 where he worked in the areas of artificial intelligence (AI), neuroimaging and network science. His PhD thesis was nominated for the best thesis award at the School of Computer Science and Engineering at NTU, SIngapore. He pursued a bachelor’s in engineering (Computer Science) from Punjab Engineering College, Chandigarh, where he went for exchange programs to Carnegie Mellon University (US), and Ben Gurion University of the Negev (Israel). He was awarded with the Institute Color in recognition of his research activities during his UG program.
- Coping with irrational identity issues using random ideals
Abstract:Speaker: Prof. Nikhil Balaji
Date: 2022-03-10 Time: 12:00:00 (IST) Venue: Teams Identity testing is a fundamental problem in algorithmic algebra. In particular, identity testing in number fields has been much studied in relation to solving systems of polynomial equations, polynomial identity testing, and decision problems on matrix groups and semi groups, among many other problems. Among number fields, cyclotomic fields, i.e., those generated by roots of unity, play a particularly important role. I'll introduce the problem of identity testing integers in cyclotomic fields and present efficient algorithms for some special cases. As an application, we will see how cyclotomic identity testing can yield a simple parallel algorithm for testing equality of compressed strings.No background in number theory or computational complexity will be assumed for this talk.Based on joint work with Sylvain Perifel, Mahsa Shirmohammadi and James Worrell.
2020 talks
- Communication complexity based approaches to formula lower bounds and lifting theorems
Abstract:Speaker: Dr Sajin Koroth, Simon Fraser University, Canada
Date: 2020-11-28 Time: 12:00:00 (IST) Venue: Microsoft teams Despite the tremendous progress in the last couple of decades, many of the fundamental questions about the nature of computation remain open. This lack of progress has been explained by limitations of well-known techniques that drove progress in the past. The main focus of complexity theory in recent times has been the search for tools and techniques that avoid such barriers. This talk is motivated by the following (seemingly different) fundamental questions that remain wide open:
• Are there natural problems that are poly time solvable but cannot be solved efficiently by parallel computation?
• What connections exist between optimality in weak and strong models of computation?
We make progress on both the problems using connections and tools from communication complexity and information theory that avoid the known barriers. The first question is central problem in complexity theory and computational problems with ``efficient'' parallel algorithms are an important computational class in the theory and practice of algorithms. Most of the linear algebra can be done using efficient parallel algorithms, yet a fundamental problem like linear programming (solvable in poly-time) is not known to have efficient parallel algorithms. We make progress on the first question, known as the $P neq NC^1$ problem, by studying a related conjecture called KRW conjecture (proposed by Karchmer, Raz, and Wigderson in the '90s).
KRW conjecture reduces the $P neq NC^1$ problem to studying a fundamental operation of Boolean functions called block-composition in the framework of communication complexity. We prove a variant of the KRW conjecture and open up a path to solve $P neq NC^1$ via a restricted version of the KRW conjecture. Block-composition is also a central operation that underlies a fundamental approach to the second question, known as "lifting theorems". Lifting theorems, try to lift lower bounds from a weaker model of computation (usually decision trees) to a stronger model of computation (usually communication complexity). Lifting theorems have been instrumental in resolving long-standing open problems in many areas of theoretical computer science like communication complexity, circuit complexity, proof complexity, linear programming, game theory, etc. We make progress on this problem by proving new lifting theorems that work for a large class of Boolean functions. Because of the efficiency and generality of our lifting theorems, our results have found applications in many different areas including circuit complexity, proof complexity, quantum communication complexity, and query complexity. The challenges involved in studying these problems in the framework of communication are fundamentally different, and looking forward, we suggest new approaches and tools.
Bio:Sajin Koroth is a postdoctoral researcher at Simon Fraser University. His research interests are in Complexity Theory, specifically, in circuit complexity and communication complexity and the interplay between these two seemingly different areas. Earlier he was a postdoctoral fellow at the University of Haifa. During this time he attended the Simons program on Lower bounds in Computational Complexity at the University of California, Berkeley as a visiting postdoc. He obtained his Ph.D. from the Indian Institute of Technology, Madras under the guidance of Jayalal Sarma.
- Internet (Anti-)Censorship – Privacy, Anonymity and the Whole Nine Yards
Abstract:Speaker: Sambuddho Chakravarty, Asst. Prof. , IIIT Delhi
Date: 2020-11-20 Time: 12:00:00 (IST) Venue: Microsoft teams In the second decade of the century (circa the Arab Springs of 2011), the Internet is the new battlefield where wars between politicians, media, (h)activists, lawyers and the military, shape the destiny of millions of people. Historically incepted as the ARPANET, it was engineered to serve as means of communication, even in the face of calamities and wars. Political will often plays antithetical to this very attribute. For instance, countries like China, Iran and UAE use (homebrewed) firewalling infrastructure to censor web traffic -- sometimes with the pretext of preserving cultural and religious values, at other times to prevent political dissent. No wonder a large body of network censorship measurements focuses on these two countries. While such countries are inherently (constitutionally) undemocratic, ``free speech'' over the Internet is, in recent years, being regularly suppressed even in democracies like India. Such evolutions are positioned on concerns otherwise paramount to the preservation of human rights -- e.g., policing child pornography. But state control of communication channels has been abuse to silence dissent, even in India where the supreme court deems freedom of speech on the Internet a fundamental right.
In this context, it is natural to ask how free and open the Internet is and how robust it is to censorship by countries like India, that in the recent years has evolved a sophisticated censorship infrastructure.
In this talk I present our work over the years that has focussed on evolution of Indian’s Internet censorship infrastructure, how it censors traffic (and now apps.), how various ISPs implement it. Further, I also present some research efforts to evade censorship (and also Internet shutdowns/blackouts).
To begin with we consider the question of whether India might potentially follow the Chinese model and institute a single, government-controlled filter. Our research shows that would not be difficult, as the Indian Internet is quite centralized already. A few ``key'' ASes (~ 1% of Indian ASes) collectively intercept approximately 95% of paths to the censored sites we sample in our study, and also to all publicly-visible DNS servers. About 5000 routers spanning these key ASes would suffice to carry out IP or DNS filtering for the entire country; approximately 75% of these routers belong to only two private ISPs. Thereafter we conducted an extensive study (first of its kind) involving nine major ISPs of the country in terms of what kind of censorship techniques they use, what triggers them, their consistency and coverage, and how to evade them. Our results indicate a clear disparity among the ISPs, on how widely they install censorship infrastructure. We also compare our findings against those that are obtained using the Open Observatory of Network Interference (OONI) tool, the latter surprisingly has several false positives and negatives. As of now we are looking into how Indian ISPs have evolved their censorship strategies to filter TLS based traffic, along with implementing the latest mandate to block mobile apps.
While existing solutions to evade censorship include proxies, VPNs, Tor have been designed primarily for web, while other applications like VoIP (real-time voice) are mostly ignored. As a part of our research we have extensively explored the feasibility of transporting real-time voice (mostly UDP) over Tor (that primarily supports TCP). Prior research deemed Tor to be unsuitable for such purposes. In our research we tried to identify how the interplay of network attributes (delay, jitter, bandwidth etc.) impact performance of VoIP. To our surprise the belief established from prior research seems unfounded.
However, all such solutions that rely on proxies are prone to being filtered by the ISPs, as these end-points are easily discoverable. Futuristic solutions like Decoy Routing, that rely on routers that could double as “smart proxies”, are resilient to such filtering. They have hitherto relied mostly on commodity servers, and involve wide scale traffic observation, inadvertently posing a threat to the privacy of users who do not require such services. To that end, we devise a SDN based DR solution, SiegeBreaker, that not only performs at line rates (comparable to native TCP) but also does not require inspection of all network flows, thus preserving the privacy of oblivious users. However, the deployability of such solutions remains a challenge, as it requires support from major top-tier ISPs.
A third alternative, combining the best of both the above solutions, involves tunnelling Interent traffic over that of various (semi-)real time applications, e.g. Instant Messaging (IM). To that end, we designed and tested a scheme, Camoufler, that utilizes IM channels textit{as-is} for transporting traffic. The scheme provides unobservability and good QoS, due to its inherent properties, such as low-latency message transports. Moreover, unlike Decoy Routing, it does not pose new deployment challenges. Performance evaluation of Camoufler, implemented on five popular IM apps indicate that it provides sufficient QoS for web browsing. E.g., the median time to render the homepages of Alexa top-1k sites was recorded to be about 3.6s, when using Camoufler implemented over Signal.
Bio:Sambuddho Chakravarty works as an Asst Prof. at the Department of Computer Science and Engineering Department of Indraprastha Institute of Information Technology (IIIT Delhi) since June 2014. He completed his PhD in Computer Science from Columbia University, New York, where he worked at the Network Security Lab (NSL) and was advised by Prof. Angelos D. Keromytis. His research is broadly related to network security and more specifically related to network anti-censorship, counter surveillance, network anonymity and privacy (and all problems revolving around such systems e.g. network measurements, infrastructure etc.). He heads a small research lab at IIIT Delhi that involves ten students (mostly PhDs and B.tech students) and collaborates actively with other networks and systems security researchers in India and abroad.
- Fairness in Unsupervised Machine Learning: From Political Philosophy to Data Science Algorithms
Abstract:Speaker: Dr. Deepak Padmanabhan, Queen's Univ. Belfast (UK)
Date: 2020-02-28 Time: 12:00:00 (IST) Venue: Bharti Building #501 Data science has long considered learning better models and making better decisions based on existing data. With data science systems penetrating into every aspect of daily life, whether it be navigation systems that influence which route you pick for your commute or personalized news delivery systems that decide what kind of news you would like to read, it is increasingly critical to consider questions on fairness. For example, would you be more prone to pre-emptive checks at the hands of an AI system if you were of a different gender, race or are a native of a different region?. There are various kinds of biases that data driven systems can internalize based on the data they train over, the assumptions they use, and the optimization that they model. Two prominent doctrines are those of disparate treatment and disparate impact, and have been subject to much study in the context of classification systems. The scenario of unsupervised learning poses a deeper challenge, where the detection of biases is often trickier given the absence of labels in the data. This talk will introduce doctrines of fairness from political philosophy, and cover streams of research on incorporating notions of fairness within retrieval and clustering tasks, including a recent work by the speaker on fairness in clustering. This will also briefly outline some fairness research directions which may be of interest for future data science research.
Bio:Dr. Deepak Padmanabhan holds a faculty position in Computer Science at Queen's University Belfast (UK) and is an adjunct faculty member at IIT Madras. He received his B.Tech from CUSAT and his M.Tech and PhD from Indian Institute of Technology Madras, all in Computer Science. His current research interests include data analytics, natural language processing, machine learning, similarity search and information retrieval. Deepak has published over 70 research papers across major venues in information and knowledge management. Before, he was a researcher at IBM Research - India for ten years. His work has led to twelve patents from the USPTO. A Senior Member of the IEEE and the ACM and a Fellow of the UK Higher Education Authority, he is the recipient of the INAE Young Engineer Award 2015, an award recognizing scientific work by researchers across engineering disciplines in India. He has also authored multiple books on various areas of data science. He may be reached at deepaksp@acm.org (preferred) or d.padmanabhan@qub.ac.uk
- Drones as On-Demand Infrastructure for Next-Generation Wireless Networks
Abstract:Speaker: Dr. Ayon Chakraborty, Researcher at NEC Laboratories in Princeton, New Jersey
Date: 2020-02-26 Time: 12:00:00 (IST) Venue: Bharti Building #501 Ubiquitous connectivity and sensing are among the grand challenge problems for next generation wireless systems, 5G and beyond. However, limitations exist in terms of resource management or infrastructure deployment (e.g., cell towers, IoT sensors etc) that hinders the realization of true ubiquity in various contexts. Often these deployments need to be on-demand, quick and temporary in order to avoid over-provisioning and maintenance costs. Recent advances in unmanned aerial vehicle (UAV) technology or drones have the potential to change the landscape of wide-area wireless connectivity and sensing by bringing a new dimension – "mobility" to the network infrastructure itself. I will demonstrate how drones can add a great value to state-of-the-art communication and sensing systems, particularly when such infrastructure are compromised (natural disasters), in-accessible (security threats), sparingly present or non-existent (rural areas).
Our aim is to design a robust networked system that helps in effective communication and coordination among people (e.g., emergency responders) in such scenarios. In the first part of the talk, I will discuss how drones can be used as "flying" cellular base stations providing connectivity to clients on the ground. We will see how the position of the drone in 3D aerial space is critical in determining the overall connectivity to the clients located on the ground. Given the drone's limited flight time, we explore how to quickly search for a near-optimal spot for the drone to operate at, optimizing the capacity of the wireless radio access network (RAN). In the second part, we will learn how the drones can also be used to localize such mobile clients in GPS-denied environments without access to any pre-deployed localization infrastructure. Overall, our contributions enhance the 5G's vision for ubiquitous connectivity by providing on-demand support for connectivity and sensing in challenged environments.
Bio:Ayon Chakraborty works as a researcher in the Mobile Communications and Networking group at NEC Laboratories America located in Princeton, New Jersey since 2017, after finishing his PhD in Computer Science from Stony Brook University, New York. He is broadly interested in designing mobile systems that interact with and interpret the physical world, spanning both algorithm design as well as system prototyping. He has held several internship positions at Bell Laboratories, Hewlett Packard Laboratories and Huawei Technologies leading to longer term collaborative efforts. He has regularly published in reputed venues for systems and networking including Infocom, NSDI, Mobicom, CoNext etc and was nominated for the best paper award at ACM Sigcomm IMC 2014. In 2011, Ayon finished his B.E. in Computer Science and Engineering from Jadavpur University, Kolkata where he was awarded the department gold medal/best project award. He is also a recipient of the DAAD WISE fellowship in 2010.
- So you want to do a Startup?
Abstract:Speaker: Dinesh Chahlia, Google
Date: 2020-02-25 Time: 12:00:00 (IST) Venue: Bharti Building #501 India is a thriving startup ecosystem. Everyone around you is making it big as an entrepreneur. You are tempted to do something big as well but don't know where to start. You are wondering whether you should do the startup now or take that amazing offer with an established company and wait till you achieve financial security. You are not sure what is left to disrupt anymore. Should you do something with Artificial Intelligence? How about systems infrastructure? Is it possible to solve real world problems and still make big money? I would love to share my thoughts on the above.
Bio:Dinesh is an alumni of the CSE Deptt at IIT Delhi (2002 batch).
Dinesh comes with an illustrious career spanning 18 years in the software industry. He began his career working with startups developing novel and disruptive solutions in India and later moved on to work with tech giants such as Microsoft, Netflix and Google.
He is currently in India, advising early stage startups as an Advisor in Residence for a Google initiative called Google for Startups. The goal of the program is to level the playing field for startups across the world by connecting them with Google experts, entrepreneurs and advisors across the world. He has chatted with over 30 startups in India at various stages over the last few weeks and is very excited to be working with 6-8 out of those at a deeper engagement level.
Dinesh is an engineering leader at Google Cloud, spearheading several highly critical and top priority initiatives for the Google Cloud Platform. Previously, he has led various projects from ideation to realization in the areas of security intrusion detection, recommendation engines and fraud detection at Microsoft and Netflix. He was the Head of Development for Auptyma, India where he led the development of a low overhead,cross-platform JVM profiler that was acquired by Oracle in 2007.
Dinesh emphasizes on fostering the culture and processes that enable happy and productive teams. He has a passion for cloud computing and applying artificial intelligence to solve real-world problems. He is active in the startup community in the Seattle area, where he is helping budding entrepreneurs validate ideas and create execution strategies. - Privacy from the prying eyes: Protecting privacy of social data from external and internal actors
Abstract:Speaker: Prof. Mainack Mondal, IIT Kgp
Date: 2020-02-21 Time: 12:00:00 (IST) Venue: Bharti Building #501 Today millions of users are uploading billions of pieces of (often personal) text, pictures or videos via large-scale networked systems (e.g., online social media sites and cloud storage platforms). As a result, designing and building private and secure access-management systems for this online data has become a widespread and an extremely important research problem.
To that end, our research aims to design and build data-driven usable, private and secure systems to assist users in managing access to their online data. In this talk, I will focus on online social media sites (OSMs) like Facebook and Twitter, which are used by millions of users and plagued by frequent well-publicized privacy and security issues. In OSMs our work considers protecting privacy of social data against actors who are external to social media platform as well as internal actors. I will first cover our work on protecting privacy from a set of external actors – large-scale third-party data aggregators like Spokeo who crawl and aggregate large-scale data from unsuspecting social media users. Specifically, I will present Genie -- a system we built to protect privacy of social data by limiting large-scale data aggregators. Next, I will move to our recent work on protecting privacy of old archival social data from a set of internal actors -- unwanted OSM users. Today the social media platforms often force users to choose a privacy setting while uploading data. However, changes in users' lives and relationships, as well as social media platforms themselves, can cause mismatches between a post's active privacy setting and the desired setting. To that end, I will present how we attacked this problem through a combination of a user study and the development of automated inference to identify potentially mismatched privacy settings. Finally, I will touch upon our work on "deletion privacy"; This research direction deals with understanding the content removal practices in social media, privacy concerns associated with it and protecting privacy of the deleted content. I will conclude this talk with a brief overview of my current research agenda and ongoing work on building usable, secure and private access-management systems for online data.
Bio:Dr. Mainack Mondal is an assistant professor of Computer Science at IIT Kharagpur. He completed his Ph.D. from Max Planck Institute for Software Systems (MPI-SWS), Germany in 2017. Prior to joining IIT Kharagpur he was a postdoctoral researcher in University of Chicago and Cornell Tech.
Mainack is broadly interested about incorporating human factors in security and privacy, and consequently designing usable online services. Specifically, he works on developing systems which provides usable privacy and security mechanisms to online users while minimizing system abuse. His work has led to papers in ACM's CCS, PETS, AAAI's ICWSM, Usenix's SOUPS, ACM's CSCW, ACM's CoNExt and Usenix's EuroSys among others. His work also received distinguished paper award in Usenix's SOUPS. - Towards automated debugging and optimization
Abstract:Speaker: Dr. Abhilash Jindal, Co-founder and CTO of Mobile Enerlytics, San Francisco, CA
Date: 2020-02-20 Time: 12:00:00 (IST) Venue: Bharti Building #501 Debugging and optimization are largely ad-hoc manual processes taking up 35-75 percent of programmers' time costing more than 0B annually. This process becomes further exacerbated in the modern programming paradigm where programmers stand on the "shoulder of giant" software frameworks to quickly program complex systems but have a limited understanding of the intricacies of the underlying system.
In the first part of this talk, we will see how new power management APIs on smartphones seriously encumber programming. These APIs, when not used correctly, give rise to a whole new class of energy bugs called sleep disorder bugs. These bugs plague the complete smartphone software stack including apps, framework, kernel, and device drivers. I will present a taxonomy of sleep disorder bugs with precise bug definitions which enabled us to create a suite of automated tools to identify different flavors of sleep disorder bugs. I will then present an automated static analysis tool KLOCK that identified 63 sleep-induced time bugs, a subclass of sleep disorder bugs, in the Android Linux kernel.
In the second part of the talk, we will study how the traditional profilers fall short in giving actionable diagnosis for optimizing programs. As even after being presented with performance hotspots, developers still need significant manual inspection and reasoning of the source code to proceed with the remaining optimization tasks: 1. Is there a more efficient implementation? 2. How to come up with a more efficient implementation? I will present a new optimization methodology called differential profiling that automates answering these two questions. In particular, differential profiling automatically uncovers more efficient API usage by leveraging existing implementations of similar apps which are bountiful in the app marketplace. Our differential energy profiling tool, DIFFPROF employs a novel tree-matching algorithm for comparing energy profiler outputs of pairs of similar apps and found 12 energy improvements in popular Android apps.
Bio:Abhilash Jindal received B.Tech. in Electrical Engineering from IIT Kanpur. He received his Ph.D. in Computer Science from Purdue University where he researched software-based approaches for extending mobile battery life. He has published papers in top system conferences including OSDI, ATC, Eurosys, Mobisys, Sigmetrics, and Mobicom. Currently, he is commercializing his research work serving as the CTO of Mobile Enerlytics, a silicon valley startup. His research interests include mobile systems, software engineering, and operating systems.
- Social media-based epidemic intelligence
Abstract:Speaker: Dr. Aditya Joshi is a Postdoctoral Fellow at CSIRO, the national science agency of Australia.
Date: 2020-02-06 Time: 12:00:00 (IST) Venue: Bharti Building #501 Epidemic intelligence is the use of technology to detect and manage outbreaks of diseases. In this talk, I will present our work on early detection of epidemics using social media posts. This work is divided into two parts: (a) health mention classification (the use of natural language processing to detect illness reports in a social media post), and (b) health event detection (the use of time series monitoring to detect epidemics from a social media stream). For health mention classification, we propose a linguistically motivated deep learning-based architecture. For health event detection, we experiment with an asthma outbreak in Melbourne in 2016. I will end the talk with a discussion on opportunities for social media-based epidemic intelligence in India: the second mobile phone market and a richly multilingual country with a peculiar social media usage pattern.
Bio:Aditya Joshi is a Postdoctoral Fellow at CSIRO, the national science agency of Australia. He holds a joint PhD degree from IIT Bombay and Monash University, Australia. He has worked for a database MNC, an artificial intelligence startup and a government research organisation, each of which provided a valuable perspective. His teaching talks span diverse forums: NLP schools organised by IITB and IIIT-H; a tutorial at EMNLP, a leading NLP conference; and a TEDx talk. His research was covered in media stories in the Indian Express, the Times of London and the Australian. He received the best PhD thesis award from the IITB-Monash Research Academy.
- A Wakeup Call: Databases in an Untrusted Universe
Abstract:Speaker: Prof. Amr El Abbadi, University of California, Santa Barbara (UCSB)
Date: 2020-01-31 Time: 15:00:00 (IST) Venue: SIT Building #001 Once upon a time databases were structured, one size fit all and they resided on machines that were trustworthy and even when they failed, they simply crashed. This era has come and gone
as eloquently stated by Mike Stonebraker. We now have key-value stores, graph databases, text databases, and a myriad of unstructured data repositories. However, we, as a database community still cling to our 20th-century belief that databases always reside on trustworthy, honest servers. This notion has been challenged and abandoned by many other Computer Science communities, most notably the security and the distributed systems communities. The rise of the cloud computing paradigm, as well as the rapid popularity of blockchains, demand a rethinking of our naïve, comfortable beliefs in an ideal benign infrastructure. In the cloud, clients store their sensitive data in remote servers owned and operated by cloud providers. The Security and Crypto Communities have made significant inroads to protect both data and access privacy from malicious untrusted storage providers using encryption and oblivious data stores. The Distributed Systems and the Systems Communities have developed consensus protocols to ensure the fault-tolerant maintenance of data residing on untrusted, malicious infrastructure. However, these solutions face significant scalability and performance challenges when incorporated in large scale data repositories. Novel database designs need to directly address the natural tension between performance, fault-tolerance, and trustworthiness. This is a perfect setting for the database community to lead and guide. In this talk, I will discuss the state of the art in terms of data management in malicious, untrusted settings, its limitations, and potential approaches to mitigate these shortcomings. As examples, I will use cloud and distributed databases that reside on untrustworthy malicious infrastructure and discuss specific approaches for standard database problems like commitment and replication. I will also explore blockchains, which can be viewed as asset management databases in untrusted infrastructures.
Bio:Amr El Abbadi is a Professor of Computer Science at the University of California, Santa Barbara. He received his B. Eng. from Alexandria University, Egypt, and his Ph.D. from Cornell University. His research interests are in the fields of fault-tolerant distributed systems and databases, focusing recently on Cloud data management and blockchain-based systems. Prof. El Abbadi is an ACM Fellow, AAAS Fellow, and IEEE Fellow. He was Chair of the Computer Science Department at UCSB from 2007 to 2011. He has served as a journal editor for several database journals, including, The VLDB Journal, IEEE Transactions on Computers and The Computer Journal. He has been Program Chair for multiple database and distributed systems conferences. He currently serves on the executive committee of the IEEE Technical Committee on Data Engineering (TCDE) and was a board member of the VLDB Endowment from 2002 to 2008. In 2007, Prof. El Abbadi received the UCSB Senate Outstanding Mentorship Award for his excellence in mentoring graduate students. In 2013, his student, Sudipto Das received the SIGMOD Jim Gray Doctoral Dissertation Award. Prof. El Abbadi is also a co-recipient of the Test of Time Award at EDBT/ICDT 2015. He has published over 300 articles in databases and
distributed systems and has supervised over 35 Ph.D. students. - Fairness in Algorithmic Decision Making
Abstract:Speaker: Dr. Abhijnan Chakraborty is a Post-doctoral Researcher at the Max Planck Institute for Software Systems (MPI-SWS), Germany.
Date: 2020-01-28 Time: 12:00:00 (IST) Venue: Bharti Building #501 Algorithmic (data-driven) decision making is increasingly being used to assist or replace human decision making in domains with high societal impact, such as banking (estimating creditworthiness), recruiting (ranking job applicants), judiciary (offender profiling), healthcare (identifying high-risk patients who need additional care) and journalism (recommending news-stories). Consequently, in recent times, multiple research works have uncovered the potential for bias (unfairness) in algorithmic decisions in different contexts, and proposed mechanisms to control (mitigate) such biases. However, the emphasis of existing works has largely been on fairness in supervised classification or regression tasks, and fairness issues in other scenarios remain relatively unexplored. In this talk, I will cover our recent works on incorporating fairness in recommendation and matching algorithms in multi-sided platforms, where the algorithms need to fairly consider the preferences of multiple stakeholders. I will discuss the notions of fairness in these contexts and propose techniques to achieve them. I will conclude the talk with a list of open questions and directions for future work.
Bio:Abhijnan Chakraborty is a Post-doctoral Researcher at the Max Planck Institute for Software Systems (MPI-SWS), Germany. He obtained his PhD from the Indian Institute of Technology (IIT) Kharagpur under the supervision of Prof. Niloy Ganguly (IIT Kharagpur) and Prof. Krishna Gummadi (MPI-SWS). During PhD, he was awarded the Google India PhD Fellowship and the Prime Minister's Fellowship for Doctoral Research. Prior to joining PhD, he spent two years at Microsoft Research India, working in the area of mobile systems. His current research interests fall under the broad theme of Computing and Society, covering the research areas of Social Computing, Information Retrieval and Fairness in Machine Learning. He has authored several papers in top-tier computer science conferences including WWW, KDD, AAAI, CSCW, ICWSM and MobiCom. His research works have won the best paper award at ASONAM'16 and best poster award at ECIR’19. He is one of the recipients of internationally competitive research grant from the Data Transparency Lab to advance his research on fairness and transparency in algorithmic systems. More details about him can be found at
https://people.mpi-sws.org/~achakrab - Scalable algorithms for rapidly advancing DNA sequencing technologies
Abstract:Speaker: Dr. Chirag Jain is currently a postdoctoral fellow with Dr. Adam Phillippy in the Genome Informatics Section at the National Institutes of Health (NIH), USA.
Date: 2020-01-27 Time: 12:00:00 (IST) Venue: Bharti Building #501 Genomics continues to have an immense impact in life sciences, from elucidating fundamental genetic mechanisms to providing new keys to understand diseases. Catalyzed by breakthroughs in genomic technologies, high-throughput DNA sequencing has become a major generator of data, catching up with the most prominent big-data
applications such as astronomy and social media. As a result, it has become challenging to design scalable DNA sequence algorithms with suitable accuracy guarantees.
Bio:Dr. Chirag Jain is currently a postdoctoral fellow with Dr. Adam Phillippy in the Genome Informatics Section at the National Institutes of Health (NIH), USA. He got his doctorate in Computational Science at Georgia Tech under the supervision of Prof. Srinivas Aluru, and bachelors in Computer Science from IIT Delhi. He is a recipient of the Best GPU
Paper Award at IEEE HiPC’14, Best Poster awards at RECOMB’19, CRIDC’19, IITD-OpenHouse’14, and Reproducibility-Initiative Award at ACM Supercomputing’16. - Security and Privacy of Connected Autonomous Vehicles
Abstract:Speaker: Dr. Vireshwar Kumar, postdoc at Purdue University
Date: 2020-01-22 Time: 00:00:00 (IST) Venue: Bharti-101 The upcoming smart transportation systems which consist of connected autonomous vehicles, are poised to transform our everyday life. The sustainability and growth of these systems to their full potential will significantly depend on the robustness of these systems against security and privacy threats. Unfortunately, the communication protocols employed in these systems lack mainstream network security capabilities due to energy constraints of the deployed platforms and bandwidth constraints of the communication medium. In this talk, I will present the results of my efforts in anatomizing the two vital communication protocols employed in the smart transportation: (1) vehicle-to-everything (V2X) communication protocol which is utilized to facilitate wireless communication among connected vehicles, and (2) controller area network (CAN) protocol which is utilized within an autonomous vehicle to enable real-time control of critical automotive components including brakes. For each of these two protocols, I will first describe the inquisitive approach which led to the discovery of the new security vulnerabilities. Then, through the experiments on real-world systems, I will demonstrate how these vulnerabilities can be exploited to launch malicious attacks which evade the state-of-the-art defense mechanisms employed in these systems. I will conclude the talk by discussing novel countermeasures which are required to mitigate these fundamental vulnerabilities and prevent their exploitation.
Bio:Dr. Vireshwar Kumar is a Postdoctoral Research Associate in the Department of Computer Science at Purdue University. Vireshwar earned his B.Tech. in Electrical Engineering at Indian Institute of Technology Delhi in 2009, and Ph.D. degree in Computer Engineering at Virginia Tech in 2016. He was the recipient of the outstanding Ph.D. student award by the Center for Embedded Systems for Critical Applications at Virginia Tech. He also had a short stint as a Project Assistant in the Department of Electrical Communication Engineering at Indian Institute of Science in 2010. His research interests include discovering and mitigating security vulnerabilities in the communication protocols employed in cyber-physical systems, e.g., smart home, smart
transportation and smart city. Vireshwar’s research work has featured in top-tier security venues including ACM Conference on Computer and Communications Security (CCS) and IEEE Transactions on Information Forensics and Security (TIFS). He has also served on the TPC of flagship conferences including IEEE Conference on Communications and Network Security (CNS) and IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN). - Redesigning how networks work to make the net"work"
Abstract:Speaker: Dr. Praveen Tammana, Postdoctoral researcher at Princeton University
Date: 2020-01-16 Time: 12:00:00 (IST) Venue: Bharti Building #501 Solving network-management problems in real-time becomes increasingly difficult with a human in the middle and ever-growing demands of modern applications for high performance, high availability, and high security. An alternative for making network management easier while meeting the demands is to work on an ambitious goal of self-managing networks that are able to manage themselves with minimal human involvement.
Practically deployable self-managing networks heavily rely on fine-grained telemetry data to understand what is going on in the network (visibility) and then make appropriate network management decisions. However, it is extremely challenging and technically expensive to build a telemetry system that provides necessary visibility; mainly because of the large resource overhead in monitoring and collecting the telemetry data from massive networks that can have up to millions of links and thousands of high-speed switches. Today, due to limited resources, network infrastructure offers poor visibility, as a consequence, operators have to do a lot of manual work or a lot of estimations, which oftentimes lead to inaccurate decisions.
In this talk, I will highlight a scalable network telemetry system, SwitchPointer, which obtains high visibility by making architectural changes across the network entities (e.g., switches, servers) in a large-scale data center. The main challenge is how to effectively use limited resources at switches and obtain high visibility. I will discuss in detail how SwitchPointer addresses this challenge and enables taking fast and accurate management decisions.
Bio:Praveen Tammana is currently a Postdoctoral researcher at Princeton University. His research interests are in Systems and Networking. He develops scalable, fast, and easy-to-use real-world networked systems. His recent work focuses on two aspects of network management: network telemetry and traffic engineering. Prior to Princeton, he received his Ph.D. from The University of Edinburgh, UK, and his Master's degree from IIT-Madras, India. He has over three years of industrial experience working with Intel technology and Juniper Networks, at Bangalore, and Cisco Systems, San Jose, USA.
- Text is not Text: Challenges in deep text understanding in professional domains
Abstract:Speaker: Vijay Saraswat, Global Head of AI R&D, Goldman Sachs
Date: 2020-01-14 Time: 16:30:00 (IST) Venue: SIT Building #001 Thanks to Big Data, Compute, Code (and, surprisingly, People), NLU research has entered a golden period. Hitherto, much of the impetus for this work has come from the desire to computationally understand "mass content" -- content from the web, social media, news sources. Here, relatively shallow meaning extraction techniques have worked reasonably well, without needing to use linguistically motivated, deep, constrained-based NL systems such as LFG and HPSG.
Significant challenges arise, however, when one starts to work with text in professional domains (e.g., financial or legal). Here documents such as regulations, contracts, agreements (e.g., loan, credit, master service), financial prospectuses, company and analyst reports must be addressed.
A contract (e.g. commercial line of credit) may involve multiple agreements with (typically, overriding) amendments, negotiated over many years. References are used at multiple semantic levels, and written using genre-specific conventions (e.g., |Casualty Event pursuant to Section 2.05(b)(ii)(B)|, |the meaning specified in Section 5.14|). Documents (e.g. EU regulations) may contain editing instructions that specify amendments to previous documents by referencing their clauses and supplying (quoted) replacement text.
Such documents typically contain highly complex text (very low Flesch scores), with single sentences spreading over multiple paragraphs, with named sub-clauses. They may use specific conventions (e.g. parenthetical structures) to present parallel semantic propositions in a syntactically compact way. They may use technical terms, whose meaning is well-understood by professionals, but may not be available in formalized background theories. Moreover, documents typically contain definitional scopes so that the same term can have various meanings across documents.
Further, documents usually define hypothetical or potential typical events (e.g. |events of default|), rather than actual (or fake) concrete events (e.g. |Barack Obama was born in Kenya|); text may be deontic, not factual. Text may specify complex normative definitions, while carving out a series of nested exceptions. It may include sophisticated argumentation structures (e.g. about company valuations) that capture critical application-specific distinctions. Ironically, in some cases we see rather contorted text (e.g. defining contingent interest rates) which is essentially a verbalization of mathematical formulas.
In short: professional text has an enormously rich structure, refined over centuries of human business interactions. This structure is distant from the news (WSJ, NYT) corpora used for "broad domain" NL research, and even the "English as a Formal Language" approach of traditional linguists.
We outline a long-term research agenda to computationalize such documents. We think of language processors as compilers, operating on the input document at varying levels of abstraction (abstract syntax tree, intermediate representation) and using a variety of techniques (partial evaluation, abstract interpretation) to generating meaning representations (rather than object code), intended primarily for use with back-end reasoners. We hypothesize the need to extend the highly sophisticated, large-parameter pattern matching characteristic of today's deep learning systems with linguistically rigorous analyses, leveraging logical representations.
Bio:Vijay Saraswat graduated from IIT Kanpur in 1982 with a B Tech in Electrical Engineering, and from Carnegie-Mellon University in 1989 with a PhD in Computer Science. Over thirty years, he has been a Member of the Research Staff at Xerox PARC, a Technology Consultant at AT&T Research, and a Distinguished Research Staff Member and Chief Scientist at IBM TJ Watson Research Center.
Vijay's research interests span a number of areas in Computer Science, across AI, logic and programming systems. He is particularly known for his work on concurrent constraint programming, (with Mary Dalrymple and colleagues) on "glue semantics", and on the X10 programming language for high performance computation. His work has won numerous awards.
Vijay joined Goldman Sachs in 2017 to help found corporate R&D. He currently leads the AI R&D work at GS, with team members in New York, Bangalore, Frankfurt, Hong Kong and other locations. The team is focused on natural language understanding, knowledge extraction, representation and reasoning. The team is looking to establish relationships with key academic and industrial partners, with engagements geared towards creation of public data-sets and associated publications. - Attention and its (mis)interpretation
Abstract:Speaker: Danish Pruthi, CMU
Date: 2020-01-14 Time: 14:00:00 (IST) Venue: SIT Building #001 Attention mechanisms are ubiquitous components in neural architectures applied to natural language processing. In addition to yielding gains in predictive accuracy, attention weights are often claimed to confer interpretability, purportedly useful for explaining why a model makes its decisions to stakeholders. In the talk, I will discuss a simple technique for training models to produce deceptive attention masks, calling into question the use of attention for interpretation. Further the talk will characterize the cost of deception (in terms of accuracy), and shed light into alternative mechanisms that the models use to cheat.
Bio:Danish Pruthi is a Ph.D. student at School of Computer Science in Carnegie Mellon University. Broadly, his research aims to enable machines to understand and explain natural language phenomenon. He completed his bachelors degree in computer science from BITS Pilani, Pilani in 2015. He has also spent time doing research at Facebook AI Research, Microsoft Research, Google, and Indian Institute of Science. He is a recipient of the Siebel Scholarship, and CMU Presidential Fellowship.
- Models for Human-AI Decision Making
Abstract:Speaker: Prof. Ambuj Singh, University of California, Santa Barbara (UCSB)
Date: 2020-01-14 Time: 15:30:00 (IST) Venue: Bharti Building #101 Teams of the future will likely consist of humans and AI agents. To this purpose, we conducted experiments to explore how such teams integrate their individual decisions into a group response. We propose a set of models to explain team decision making using appraisal dynamics, Prospect Theory, Bayes rule, and eigenvector centrality. The first two models encode how individuals in a group appraise one other's expertise and the risk-rewards of their opinions. The second two models encode how groups integrate their members' preferences to form a group decision. Decision-making in the experiments proceeds consists of two sequential tasks: a first task in which the teams decide to report one of the presented options to a multiple-choice question or to choose one of the agents. If the teams decide to use an agent, the second decision-making task consists of integrating the agent's response with their previous responses and reporting an answer. Our findings suggest that the proposed models explain decision making in teams. Furthermore, the teams exceed the performance of the best human member most of the time and the AI agents in the first decision task but not so in the second decision task.
Bio:Ambuj K. Singh is a Professor of Computer Science at the University of California, Santa Barbara, with part-time appointments in the Biomolecular Science and Engineering Program and the Technology Management Program. He received a B.Tech. degree from the Indian Institute of Technology, Kharagpur, and a Ph.D. degree from the University of Texas at Austin. His research interests are broadly in the areas of network science, machine learning, and bioinformatics. He has led a number of multidisciplinary projects including UCSB's Interdisciplinary Graduate Education Research and Training (IGERT) program on Network Science funded by the National Science Foundation (NSF). He is currently directing a Multidisciplinary University Research Initiative (MURI) on the Network Science of Teams. He has graduated over 25 Ph.D. students over his career.
- A Regularization view of Dropout in Neural Networks
Abstract:Speaker: Ambar (PHD)
Date: 2020-01-13 Time: 12:00:00 (IST) Venue: SIT Building #001 Dropout is a popular training technique used to improve the performance of Neural Networks. However, a complete understanding of the theoretical underpinnings behind this success remains elusive. In this talk, we will take a regularization view of explaining the empirically observed properties of Dropout. In the first part, we will investigate the case of a single layer linear neural network with Dropout applied to the hidden layer, and observe how the Dropout algorithm can be seen as an instance of Gradient Descent applied to a changing objective. Then we will understand how training with Dropout can be seen to be equivalent to adding a regularizer to the original network. With these tools we would be able to show that Dropout is equivalent to a nuclear-norm regularized problem, where the nuclear-norm is taken on the product of the weight matrices of the network. Inspired by the success of Dropout, several variants have been proposed recently in the community. In the second part of the talk, we will analyze some of these variants (DropBlock and DropConnect), and obtain theoretical reasons for their success over vanilla Dropout. Finally, we will end with a unified theory to analyze Dropout variants, and understand some of the implications.
Bio:Ambar is a PhD student in the Computer Science Department at the Johns Hopkins University. He is advised by René Vidal, and affiliated with the Mathematical Institute for Data Science and the Vision Lab at JHU. Previously he has obtained his Bachelor's degree in Computer Science from IIIT Delhi. His current research interest lies in the theory of deep learning, specifically, to theoretically understand the properties induced by common deep learning techniques on the optimization of deep architectures. He is currently working on understanding the regularization properties induced by common tricks used in training DNNs. His other interests include understanding adversarial examples generated for computer vision systems. Ambar is MINDS Data Science Fellow and his research is also supported by the IARPA DIVA and the NSF Optimization for Deep Learning grants.
- Top Ten Challenges to Realizing Artificial Intelligence (AI)
Abstract:Speaker: Dr. Ravi Kothari, Ashoka U.
Date: 2020-01-13 Time: 16:00:00 (IST) Venue: Bharti Building #501 The many successes of deep-learning notwithstanding, there remain fundamental impediments to realizing true Artificial Intelligence (AI). In this widely accessible talk, we discuss the top ten challenges to realizing true AI and highlight some of the ongoing efforts towards resolving these challenges.
Bio:Dr. Ravi Kothari started his professional career as an Assistant Professor in the Department of Electrical and Computer Engineering of the University of Cincinnati, OH, USA where he later became a tenured Associate Professor and Director of the Artificial Neural Systems Laboratory. After about 10 years in academia, he joined IBM Research and held several positions over the years including that of Chief Scientist of IBM Research, India and the Global Chief Architect of the IBM-Airtel outsourcing account where he introduced the first-ever stream based processing of telecom data (Airtel is one of the world's largest full service telecom providers and the Chief Architect is responsible for the IT strategy, design and realization across more than 15 countries). He also became IBM's first Distinguished Engineer from India. After about 15 years in IBM, he joined Accenture for a short stint as a Fellow, Technical Labs prior to returning to academia as a Professor of Computer Science at Ashoka University.
Dr. Kothari's expertise lies in Machine Learning, Pattern Recognition, AI, Data Mining, Big Data and other data-driven initiatives. His present research focuses on multiple aspects of Artificial Intelligence including "creative" machines.
Dr. Kothari has served as an Associate Editor of the IEEE Transactions on Neural Networks, IEEE Transactions on Knowledge and Data Engineering, Pattern Analysis and Applications (Springer) as well as on the program committees of various conferences. He has been an IEEE Distinguished Visitor for many years, a member of the IBM Academy of Technology and the IBM Industry Academy. He was a recipient of the Gerstner Award (IBM's highest team award), the Best of IBM award (IBM's highest individual award) and is a fellow of the Indian National Academy of Engineering. - Parameterized Complexity of Network Design Problems
Abstract:Speaker: Dr. Pranabendu Misra, postdoctoral fellow at the Max Planck Institute for Informatics, Saarbrucken, Germany
Date: 2020-01-06 Time: 12:00:00 (IST) Venue: Bharti-501 Network Design Problems, which concern designing minimum cost networks that satisfy given set of ``connectivity constrains'', are very well studied in computer science and combinatorial optimization. Almost all these problems are NP-hard, and a number of results are known about them in the realm of approximation algorithms. Parameterized Complexity is a different framework for dealing with computational intractability, where typically we try to design fast algorithms to solve the problem on those instances which admit a ``small cost solution''. In this talk we will look at some recent results on the parameterized complexity of network design problems, and future directions for research.
Bio:Pranabendu Misra is a postdoctoral fellow at the Max Planck Institute for Informatics, Saarbrucken, Germany.Before that, he was a researcher at the Department of Informatics, University of Bergen, Norway.He obtained his PhD in Computer Science from the Institute of Mathematical Sciences, Chennai, India, advised by Saket Saurabh.He did his undergraduate studies at the Chennai Mathematical Institute in Mathematics and Computer Science.His primary research interests lie in graph theory and algorithms focusing on Parameterized Complexity.He has also worked on problems in approximation algorithms, matroids, fault tolerant graphs, matching under preferences, de-randomization.
- Fault-Tolerant Reachability and Strong-connectivity Structures
Abstract:Speaker: Dr. Pranabendu Misra, postdoctoral fellow at the Max Planck Institute for Informatics, Saarbrucken, Germany
Date: 2020-01-06 Time: 12:00:00 (IST) Venue: Bharti Building #501 Network Design Problems, which concern designing minimum cost networks that satisfy given set of ``connectivity constrains'', are very well studied in computer science and combinatorial optimization. Almost all these problems are NP-hard, and a number of results are known about them in the realm of approximation algorithms. Parameterized Complexity is a different framework for dealing with computational intractability, where typically we try to design fast algorithms to solve the problem on those instances which admit a ``small cost solution''. In this talk we will look at some recent results on the parameterized complexity of network design problems, and future directions for research.
Bio:Pranabendu Misra is a postdoctoral fellow at the Max Planck Institute for Informatics, Saarbrucken, Germany.
Before that, he was a researcher at the Department of Informatics, University of Bergen, Norway.
He obtained his PhD in Computer Science from the Institute of Mathematical Sciences, Chennai, India, advised by Saket Saurabh.
He did his undergraduate studies at the Chennai Mathematical Institute in Mathematics and Computer Science.
His primary research interests lie in graph theory and algorithms focusing on Parameterized Complexity.
He has also worked on problems in approximation algorithms, matroids, fault tolerant graphs, matching under preferences, de-randomization. - Congruence Closure Algorithms for Uninterpreted and Interpreted Symbols
Abstract:Speaker: Prof. Deepak Kapur
Date: 2020-01-02 Time: 15:00:00 (IST) Venue: Bharti Building #501 Algorithms for computing congruence closure and conditional congruence closure of conditional ground equations over uninterpreted and interpreted symbols are presented. The algorithms are based on a recently developed framework for computing (conditional) congruence closure by abstracting nonflat terms by constants as proposed first in Kapur's congruence closure algorithm (RTA97). This framework is general and flexible. It is used also to develop (conditional) congruence closure algorithms for cases when associative-commutative function symbols can have additional properties including idempotency, nilpotency and/or have identities, as well as their various combination. It also works for Horn properties including extensionality of function symbols. The concept of a Horn closure of ground equations with uninterpreted as well as interpreted symbols is proposed to decide Horn equations directly.
Bio:Professor, Department of Computer Science The University of New Mexico, Albuquerque, NM USA
- Novel and Efficient Techniques for Guaranteeing Routing and Protocol Level Deadlock Freedom in Interconnection Networks
Abstract:Speaker: Mayank Parasar (Georgia Tech)
Date: 2020-01-02 Time: 12:00:00 (IST) Venue: SIT Building #001 Interconnection networks are the communication backbone for any system. They occur at various scales: from on-chip networks between processing cores, to supercomputers between compute nodes, to data centers between high-end servers. One of the most fundamental challenges in an interconnection network is that of deadlocks. Deadlocks can be of two types: routing level deadlocks and protocol level deadlocks. Routing level deadlocks occur because of cyclic dependency between packets trying to acquire buffers, whereas protocol level deadlock occurs because the response message is stuck indefinitely behind the queue of request messages. Both kinds of deadlock render the forward movement of packets impossible leading to complete system failure.
Prior work either restricts the path that packets take in the network or provisions an extra set of buffers to resolve routing level deadlocks. For protocol level deadlocks, separate sets of buffers are reserved at every router for each message class. Naturally, proposed solutions either restrict the packet movement resulting in lower performance or require higher area and power.
In my work, I propose a new set of efficient techniques for providing both routing and protocol level deadlock freedom. My techniques provide periodic forced movement to packets in the network, which breaks any cyclic dependency of packets. Breaking this cyclic dependency results in resolving routing level deadlocks. Moreover, because of periodic forced movement, the response message is never stuck indefinitely behind the queue of request messages; therefore, my techniques also resolve protocol level deadlocks.
Bio:Mayank Parasar is a fifth-year Ph.D. candidate in the School of Electrical and Computer Engineering at Georgia Institute of Technology. He received an M.S. in Electrical and Computer Engineering from Georgia Tech in 2017 and a B.Tech. in Electrical Engineering department from Indian Institute of Technology (IIT) Kharagpur in 2013.
He works in computer architecture with the research focus on proposing breakthrough solutions in the field of interconnection networks, memory system and system software/application layer co-design. His dissertation, titled Novel and Efficient Techniques for Guaranteeing Routing and Protocol Level Deadlock Freedom in Interconnection Networks, formulates techniques that guarantee deadlock freedom with a significant reduction in both area and power budget.
He held the position of AMD Student Ambassador at Georgia Tech in the year 2018-19. He received the Otto & Jenny Krauss Fellow award in the year 2015-16. - Groebner Bases: Universality, Parametricity and Canonicity
Abstract:Speaker: Prof. Deepak Kapur,
Date: 2020-01-01 Time: 11:00:00 (IST) Venue: Bharti Building #501 The talk will integrate the concepts of a universal Groebner basis which serves as a Groebner basis for all admissible term orderings, with parametric (more popularly called comprehensive) Groebner basis which serves as a Groebner basis for all possible specializations of parameters. Three different but related approaches will be presented. First one extends Kapur's algorithm for computing a parametric Groebner basis in which along with branching based on making the head coefficient nonzero or not, branching on ordering constraints is also done in order first to choose a term that could serve as the head term. The second one is based on the Kapur, Sun and Wang's algorithm for computing comprehensive Groebner basis and system but uses a reduced universal Groebner basis to generate a universal parametric Groebner basis. The third one starts with a reduced Groebner basis using one arbitrary ordering and then generate a universal comprehensive Groebner basis by incrementally changing the orderings along with partitioning specializations. The result of these algorithm is a mega Groebner basis that works for every admissible ordering as well as for any specialization of parameters.
Bio:Professor, Department of Computer Science The University of New Mexico, Albuquerque, NM USA
2019 talks
- Randomized Algorithms in Graph Connectivity: New Takes on Old Ideas
Abstract:Speaker: Prof. Debmalya Panigrahi (Duke University)
Date: 2019-12-10 Time: 14:30:00 (IST) Venue: Bharti Building #501 In this talk, I will describe new interpretations and extensions of classic randomized algorithms in graph connectivity from the 1990s. These new insights lead to better algorithmic and combinatorial results in graph and hypergraph connectivity including min-cut algorithms, sparsification techniques, and reliability analysis. The talk will be self-contained.
Bio:Prof. Debmalya Panigrahi (Duke University)
- O(d)-competitive Convex Body Chasing
Abstract:Speaker: Prof. Anupam Gupta (Carnegie Mellon University
Date: 2019-12-10 Time: 15:45:00 (IST) Venue: Bharti Building #501 We study the problem of chasing convex bodies online: given a sequence of convex bodies K_t in R^d the algorithm must respond with points x_t in K_t in an online fashion (i.e., x_t is chosen before K_{t+1} is revealed). The objective is to minimize the total distance between successive points in this sequence. We discuss how this problem generalizes well-studied problems in online algorithms, and give an algorithm that is O(d)-competitive.
Bio:Prof. Anupam Gupta (Carnegie Mellon University
- Polynomial Pass Lower Bounds for Graph Streaming Algorithms
Abstract:Speaker: Prof. Sanjeev Khanna (Univ. of Pennsylvania)
Date: 2019-12-10 Time: 16:45:00 (IST) Venue: Bharti Building #501 We present new lower bounds that show that a polynomial number of passes are necessary for solving some fundamental graph problems in the streaming model of computation. For instance, we show that any streaming algorithm that finds a minimum s-t cut in a weighted undirected graph needs essentially quadratic space unless it makes a polynomial number of passes over the graph stream.
To prove our lower bounds, we introduce and analyze a new four-player communication problem that we refer to as the hidden-pointer chasing problem. The hidden-pointer chasing problem appears flexible enough to find other applications and is therefore interesting in its own right. To showcase this, we present an interesting application of this problem beyond streaming algorithms. Using a reduction from hidden-pointer chasing, we prove that any algorithm for submodular function minimization needs to make essentially a quadratic number of value queries to the function unless it has a polynomial degree of adaptivity.
This is joint work with Sepehr Assadi and Yu Chen.
Bio:Prof. Sanjeev Khanna (Univ. of Pennsylvania)
- Applying Predicate Detection to Discrete Optimization Problems
Abstract:Speaker: Prof. Vijay Garg, Department of Electrical & Computer Engineering , The University of Texas, Austin
Date: 2019-11-15 Time: 12:00:00 (IST) Venue: Bharti Building #501 We present a method to design parallel algorithms for constrained combinatorial optimization problems. Our method solves and generalizes many classical combinatorial optimization problems including the stable marriage problem, the shortest path problem and the market clearing price problem. These three problems are solved in the literature using Gale-Shapley algorithm, Dijkstra's algorithm, and Demange, Gale, Sotomayor algorithm. Our method solves all these problems by casting them as searching for an element that satisfies an appropriate predicate in a distributive lattice. Moreover, it solves generalizations of all these problems --- namely finding the optimal solution satisfying additional constraints called lattice-linear predicates. For stable marriage problems, an example of such a constraint is that Peter's regret is less than that of Paul. Our algorithm, called Lattice-Linear Predicate Detection (LLP) can be implemented in parallel with lock-free synchronization. It just assumes atomicity of reads and writes. In addition to finding the optimal solution, our method is useful in enumerating all constrained stable matchings, and all constrained market clearing price vectors.
Bio:Vijay Garg is a Cullen Trust Endowed Professor in the Department of Electrical & Computer Engineering at The University of Texas at Austin. He received his Ph.D. in Computer Science at the University of California at Berkeley and B. Tech. in computer science at IIT, Kanpur. His research interests are in distributed computing, discrete event systems and lattice theory. He is the author of "Elements of Distributed Computing" (Wiley, 2002), "Introduction to Lattice Theory with Computer Science Applications" (Wiley, 2015), and "Modeling and Control of Logical Discrete Event Systems" (Springer, 2012). He is an IEEE Fellow.
- Synthesis and Validation of Robust Cyber Physical Systems
Abstract:Speaker: Prof. Soumyajit Dey (http://cse.iitkgp.ac.in/~soumya/)
Date: 2019-11-14 Time: 10:30:00 (IST) Venue: Bharti Building #501 Cyber Physical Systems (CPS) define the complex interaction of sensing, control law computation and actuation in real time for complex network of physical `plants'. While control theorists have dealt with CPS modeling from a theoretical point of view, platform non-idealities play their role in determining the actual control performance. Also, platform vulnerabilities and communication interfaces can be exploited by attackers to degrade control performance of such systems. We explore the usefulness of Formal Methods tools in computing reliability guarantees for such systems. We also explore how intelligent scheduling of workloads can effect CPS performance in automotive context.
Bio:Soumyajit Dey joined the dept. of CSE, IIT Kgp in May 2013. He worked at IIT Patna as assistant professor in CSE dept. from beginning of Spring 2012 to end of Spring 2013. He received a B.E. degree in Electronics and Telecommunication Engg. from Jadavpur University, Kolkata in 2004. He received an M.S. followed by PhD degree in Computer Science from Indian Institute of Technology, Kharagpur in 2007 and 2011 respectively. His research interests include 1) Synthesis and Verification of Safe, Secure and Intelligent Cyber Physical Systems, 2) Runtime Systems for Heterogeneous Platforms.
- Investigations on Adaptive Invisible Watermarking in Transform Domain
Abstract:Speaker: Dr. Prasanth Vaidya Sanivarapu
Date: 2019-11-11 Time: 11:00:00 (IST) Venue: Bharti Building #501 In the talk I am going to highlight the research work done by me. In our research work, we focused on copyright protection and ownership identification of multimedia data. Watermarking is the best approach to overcome the challenges faced during the distribution of digital data. Digital watermarking is an effective technology of embedding data (watermark signal) into multimedia data. The authentic owner can exhibit his ownership by inserting a robust watermark in an image and later can extract it for verification. Since spatial domain methods are fragile and cannot withstand to the attacks, transform domain is utilized for embedding the watermark to provide robustness. Also, factor values are calculated dynamically in most of the methods which makes the scheme fragile. To overcome the drawbacks, we have proposed adaptive blind,
semi-blind and non-blind watermarking schemes with robustness, imperceptibility, security and fidelity. A summary of the proposed watermarking schemes is given below:
1. A blind watermarking scheme based on Binary Robust Invariant Scalable
Key points (BRISK) features, contourlet transform (CT) and QR decomposition is proposed.
2. An adaptive non-blind watermarking in wavelet domain is proposed for gray-scale images where scaling and embedding factors are calculated using the Bhattacharyya distance and kurtosis.
3. By extending the previous work, a robust blind watermarking scheme on color images is designed. To make the scheme more robust against attacks, Bhattacharyya distance is used to calculate the embedding factor values adaptively. A bit manipulation gives the scheme an additional advantage to protect the watermark from attacks.
4. A semi-blind watermarking scheme using multiple decompositions is designed. To provide security to the watermark, Arnold transform is utilized in watermark encryption by generating a secret key. The embedding is done by applying Wavelet Transform, CT, Schur decomposition and SVD in sequence and vice-versa for extraction.
5. A robust non-blind watermarking method with game-theory techniques is designed. In this method, optimal solution is provided to the decision makers. To attain the better trade-off between the embedder and the extractor an optimal solution is calculated by using Iterative Elimination of Strictly Dominant Strategies (IESDS).
6. A multipurpose color image semi-blind watermarking in wavelet domain using multiple decomposition techniques is utilized, three gray-scale watermarks are embedded in the single color image to provide better
authenticity.
7. Extending the image watermarking to video watermarking, a robust and blind watermarking for color videos using redundant wavelet transform (RDWT) and SVD is proposed.
8. In medical watermarking, the patient data is hidden into ECG signal using watermarking in transform domain
Finally, the watermarking technology can be applied by providing security and authentication to the multimedia data in different sectors.
Bio:Dr. Prasanth Vaidya Sanivarapu
- Learning the Koopman Operator for Dynamic Data
Abstract:Speaker: Prof. Chandrajit Bajaj
Date: 2019-11-11 Time: 12:00:00 (IST) Venue: Bharti Building #501 Recent work in the study of dynamic systems has focussed on data driven decomposition techniques that approximate the action of the Koopman operator on observable functions of the underlying phenomena.
In particular, the data driven method of dynamic mode decomposition (DMD) has been explored, with multiple variants of the algorithm in existence, including extended DMD, DMD in reproducing kernel Hilbert spaces, a Bayesian framework, a variant for stochastic dynamical systems, and a variant that uses deep neural networks. In this talk, I shall briefly summarize the large existing work on data driven learning of Koopman operator models, and then describe new sampling based sketching approaches (SketchyCoreSVD, SketchyCoreTucker) together with matrix-valued Kernels, to achieve a deep learning architecture for accelerated Koopman operator approximations of dynamic observable data. Examples are drawn from remote sensing, bio-medical cardiac magnetic resonance video, and time series reactive flow simulations of a single ejector combustion process.
Bio:Prof. Chandrajit Bajaj is a professor at the Department of Computer Science and Oden Institute of Engineering and Sciences, University of Texas at Austin. He is a Distinguished Alumnus of IIT Delhi, and a very well-known and respected researcher in computer graphics and vision and applications in medicine, etc.
http://www.cs.utexas.edu/~bajaj - "Solar + Storage + LED + IoT = Trillion"
Abstract:Speaker: Srinivasan Keshav, University of Cambridge
Date: 2019-11-01 Time: 16:00:00 (IST) Venue: Bharti Building #501 Recent technological advances in the areas of solar photovoltaics, Lithium-Ion-based energy storage, light emitting diodes, and the Internet of Things will substantially change the energy, electrical grid, building, and transportation sectors. Some experts estimate that these will result in economic activity of about Trillion over the next two decades. In this talk, I will touch upon the recent advances in these technical areas and speculate on their economic impacts. I will also present some of my recent results that attempt to address these
changes.
Bio:Srinivasan Keshav is the Robert Sansom Professor of Computer Science at the University of Cambridge. He received a B.Tech in Computer Science and Engineering from IIT Delhi in 1986 and Ph.D. in Computer Science from the University of California, Berkeley in 1991. He was subsequently an MTS at AT&T Bell Labs and an Associate Professor at Cornell. In 1999 he left academia to co-found Ensim Corporation and GreenBorder Technologies Inc. He has been at the University of Waterloo since 2003, holding first a Canada Research Chair and then the Cisco Chair in Smart Grid. His honours include being named an ACM and IEEE Fellow, two Test of Time awards from ACM SIGCOMM, and best paper awards at ACM SIGCOMM, MOBICOM, and eEnergy. He is the author of two graduate textbooks on computer networking and has served as Chair of ACM SIGCOMM and as Associate Dean of Graduate Studies in the Faculty of Mathematics, University of Waterloo.
- Google Research India: A Peek at Research Activities and Plans
Abstract:Speaker: Dr Manish Gupta
Date: 2019-10-31 Time: 12:00:00 (IST) Venue: Bharti Building #501 We provide an overview of research activities underway and planned at the newly established AI research lab by Google in India. Some of the fundamental problems in machine learning (ML) we are investigating include a better understanding of the limitations of deep learning systems and improving their robustness. We describe plans to conduct research at the intersection of software engineering, programming languages, and ML to make software systems built using ML methodology more robust, secure, and fair. We describe Google products like GPay and Assistant, which are being used by tens of millions of users in India, and plans to apply ML in areas like natural language understanding and social networks to improve both the unique challenges in the Indian context (e.g., code mixing, diversity of languages, dialects and accents) as well as their overall capabilities (e.g., improving the ability of the Assistant to handle conversations that require a combination of knowledge, reasoning and personalization, and improving fraud detection in GPay). We describe a critical need and an opportunity to improve health outcomes globally at lower costs, and specific challenges in countries like India of an acute shortage of doctors. We describe a convolutional neural network based solution developed by our researchers for more effective screening of diabetic retinopathy, which is being deployed at hospitals in India. Finally, we describe our AI for Social Good program through which we are addressing issues like public health, education and wildlife conservation, in partnership with NGOs.
Bio:Dr. Manish Gupta is the Director of Google Research India, a new AI research lab recently announced by Google. He holds an additional appointment as Infosys Foundation Chair Professor at IIIT Bangalore. Previously, Manish has led VideoKen, a video technology startup, and the research centers for Xerox and IBM in India. As a Senior Manager at the IBM T.J. Watson Research Center in Yorktown Heights, New York, Manish led the team developing system software for the Blue Gene/L supercomputer. IBM was awarded a National Medal of Technology and Innovation for Blue Gene by US President Barack Obama in 2009. Manish holds a Ph.D. in Computer Science from the University of Illinois at Urbana Champaign. He has co-authored about 75 papers, with more than 7,000 citations in Google Scholar, and has been granted 19 US patents. While at IBM, Manish received two Outstanding Technical Achievement Awards, an Outstanding Innovation Award and the Lou Gerstner Team Award for Client Excellence. Manish is a Fellow of ACM and the Indian National Academy of Engineering, and a recipient of a Distinguished Alumnus Award from IIT Delhi.
- Are you Happy, Angry or Sad: Identifying Emotions from Walking Using Affective and Deep Features
Abstract:Speaker: Aniket Bera, University of Maryland
Date: 2019-10-31 Time: 16:00:00 (IST) Venue: Bharti Building #501 A new data-driven model and algorithm to identify the perceived emotions of individuals based on their walking styles. Given an RGB video of an individual walking, we extract his/her walking gait in the form of a series of 3D poses. Our goal is to exploit the gait features to classify the emotional state of the human into one of four emotions: happy, sad, angry, or neutral. Our perceived emotion recognition approach uses deep features learned via LSTM on labeled emotion datasets. Furthermore, we combine these features with affective features computed from gaits using posture and movement cues. We show that our mapping between the combined feature space and the perceived emotional state provides 80.07% accuracy in identifying the perceived emotions. In addition to classifying discrete categories of emotions, our algorithm also predicts the values of perceived valence and arousal from gaits.
Bio:Dr. Aniket Bera is an Assistant Research Professor at the Department of Computer Science and the University of Maryland Institute for Advanced Computer Studies (UMIACS). Prior to this, he was a Research Assistant Professor at the University of North Carolina at Chapel Hill. He received his Ph.D. in 2017 from the University of North Carolina at Chapel Hill. His core research interests are in Computer Graphics, AI, Social Robotics, Data-Driven Crowd Simulations, Cognitive modeling: Knowledge, reasoning, and planning for intelligent characters. He is working with the Geometric Algorithms for Modeling, Motion, and Animation (GAMMA) group. His current research focuses on the social perception of intelligent agents and robotics.
Webpage: https://www.cs.umd.edu/~ab/ - An Information Session on Working with Indian Political Data by TCPD
Abstract:Speaker: Mohit Kumar (Data Scientist), Priyamvada Trivedi (Associate Director ,TCPD)
Date: 2019-10-18 Time: 15:00:00 (IST) Venue: SIT Building #001 The Trivedi Centre for Political Data (TCPD) is the product of a joint venture between the Political Science and Computer Science departments at Ashoka University. In this session, drawing from projects currently underway, we will discuss the data extraction, mining and warehousing techniques that are used to get, clean and store data on Indian politics in a scientific form. Using two web applications - SURF which is an entity resolution tool for Indian names to resolve names of candidates in order to create a unique identifier for each candidate and LokDhaba which is a portal for both Vidhan Sabha and Lok Sabha elections results post 1962 - that have been developed at the Centre, we will show how our data is disseminated in the public domain. And lastly, we will show how this data can be used to analyze and understand spatial and temporal patterns of change in Indian politics. We will end the session with a short discussion on some new projects and their associated challenges.
Bio:Mohit Kumar is a data scientist and GIS engineer. He completed his undergraduate and MS by Research from the International Institute of Information Technology (IIIT) Hyderabad in computer science engineering with specialization in spatial informatics. He has been working with TCPD since July 2017 and loves trekking. He can be contacted at mohit.kumar@ashoka.edu.in.
Priyamvada Trivedi is the Associate Director of TCPD and a Visiting Faculty in the Department of Political Science at Ashoka. She completed her undergraduate in electrical engineering from Purdue University and her PhD in political science from the University of Michigan. She has been working with TCPD since June 2018 and loves trees. She can be contacted at priyamvada.trivedi@ashoka.edu.in - Fault-Tolerant Reachability and Strong-connectivity Structures
Abstract:Speaker: Keerti Choudhary, Weizmann Institute of Science
Date: 2019-10-09 Time: 12:00:00 (IST) Venue: Bharti Building #501 Reachability and strong-connectivity are fundamental graph problems. In the static setting, both problems can be solved in O(m+n) time for any directed graph G with n vertices and m edges. However, most real-world networks are prone to failures. The failures are transient and occur for a small duration of time. Under such a setting, we would like to preprocess the graph to compute data-structures that can achieve robustness by being functional even after the occurrence of failures. In recent years, researchers have studied various questions pertaining to connectivity, distances, routing, min-cuts, etc. in the domain of fault tolerance networks.
In this talk, I will discuss the following questions:
1. Can we compute a reachability tree in just linear time, after the occurrence of a small number of failures?
2. How fast can we compute the strongly connected components, again on the occurrence of failures?
3. Can we find a certificate for these structures that remains valid even after a bounded number of failures?
Our solutions will employ some extremely basic tools like max-flows, min-cuts, and heavy-path-decomposition.
Bio:Weizmann Institute of Science
- Streaming Algorithms for Matching Problems
Abstract:Speaker: Dr. Sagar Kale
Date: 2019-09-30 Time: 12:00:00 (IST) Venue: Bharti Building #501 The advent of big data necessitated design of algorithms that could cope with it. Streaming algorithms are viable for big data because they process their input sequentially and use a small amount of working memory. In this talk, I will present my research on streaming algorithms for matching problems. Matching problems have played a significant historical role in not just combinatorial optimization specifically but computer science at large and are our benchmark problems for development and understanding of a computational model. Indeed, studying them in the streaming model has led me to fastest algorithms for some combinatorial optimization problems---streaming or otherwise.
In the maximum matching problem, edges of the input graph arrive one-by-one in the stream and the goal is to compute a matching of large size. Weighted matching is a well-studied generalization of maximum matching where the graph is edge-weighted, and, in a further generalization, a submodular function is defined on the edges. Submodular functions have received a lot of attention in theoretical computer science as well as, more recently, in machine learning.
I will discuss a reduction from submodular matching to weighted matching that also extends to streaming algorithms for very general constraints such as matroids. Then I will give an overview of how to obtain almost optimal weighted matchings in constant number of passes over the stream and also in the MapReduce framework. I will conclude with future research directions.
Bio:Sagar Kale did his Ph.D. at Dartmouth College where his advisor was Prof. Amit Chakrabarti. He is currently a postdoc at EPFL where his host is Prof. Ola Svensson. His research is mainly on streaming algorithms for combinatorial optimization problems related to matchings and matroids---which are fundamental structures in computer science.
- A hacker’s journey of exploiting your IoT Cloud Applications
Abstract:Speaker: Mr Ananda Krishna (Co-founder & CTO of Astra Security)
Date: 2019-09-24 Time: 17:15:00 (IST) Venue: Bharti Building #501 IoT devices are everywhere. From cars and ACs to monitoring devices in factories. Objects around us are increasingly being Internet-enabled. Similar to the early days of the Internet, everyone has rushed into building IoT solutions but often neglected security. Since IoT devices are connected to the internet, they are vulnerable to hacking just like any other device/servers on the Internet. In this session, we will deconstruct how hackers hack IoT solutions, common security vulnerabilities (OWASP Top 10), testing methodology, attack surface area, and what can be done to prevent it.
Bio:Co-founder & CTO of Astra Security
- Speed Scaling for Parallel and Tandem Servers
Abstract:Speaker: Rahul Vaze, TIFR
Date: 2019-09-23 Time: 16:00:00 (IST) Venue: Bharti Building #501 Scheduling jobs with adjustable server speeds with corresponding speed/power cost is referred to as the speed scaling problem. Speed scaling with multiple parallel servers has been an attractive object of interest, however, the tandem setting has been relatively less explored. In this talk, we will discuss both the parallel and the tandem speed scaling problem, in the context of competitive ratio of online algorithms with worst case guarantees.
In the parallel setting, we consider the popular shortest remaining processing time (SRPT) algorithm and show that it can achieve a constant competitive ratio. In prior work, more complicated algorithms were shown to be constant competitive.
In the tandem setting, there is a series of servers and each job has to be processed by each of the servers in sequence, and as far as we know, no work has been reported on bounding the competitive ratio in the worst case setting. When all job sizes are equal, we propose an algorithm that is shown to have a constant competitive ratio. The proposed algorithm, at all times, uses the same speed on all active servers, such that the total power consumption equals the total number of outstanding jobs. The general problem remains open.
Bio:Rahul Vaze is a Reader at the School of Technology and Computer Science at the Tata Institute of Fundamental Research, Mumbai. His research interests are in Stochastic Geometry, Wireless Networks, and Combinatorial Optimization and he is an author of a book Random Wireless Networks, CUP, 2015.
- Faster algorithms for p-norm regression
Abstract:Speaker: Sushant Sachdeva, U. Toronto
Date: 2019-09-03 Time: 16:00:00 (IST) Venue: Bharti Building #501 Linear regression in L_p-norm is a canonical optimization problem that arises in several applications, including sparse recovery, semi-supervised learning, and signal processing. Standard linear regression corresponds to p=2, and p=1 or infinity is equivalent to linear programming.
In this talk, we will discuss new algorithms for obtaining high-accuracy (1+1/poly(n))-approximate solutions to such problems. Our algorithms are based on an ‘iterative refinement’ approach for p-norms, where an algorithm for computing an approximate solution can be converted into a high-accuracy algorithm. Previously, iterative
refinement was known only for L_2-regression (or equivalently, solving linear systems).
Based on this approach, we give several new high-accuracy algorithms, including 1) an algorithm for p-norm regression by solving only m^{1/3} linear systems, 2) a provably convergent and fast iterative reweighted least squares (IRLS) algorithm for p-norm regression, and 3) an algorithm for p-norm flows in undirected unweighted graphs in almost-linear time.
This talk will be based on joint works with Deeksha Adil, Richard Peng, Rasmus Kyng, and Di Wang.
Bio:Sushant Sachdeva is a faculty member at the CS dept. at the University of Toronto. He is interested in Algorithms, and its connections to optimization, machine learning, and statistics. His recent research focus has been the design of fast algorithms for graph problems.
Before joining UofT, he was a research scientist at Google. He completed his postdoc at Yale with Dan Spielman (2016), his PhD from Princeton (2013) under Sanjeev Arora, and his BTech from IIT Bombay (2008). He is the recipient of Google Faculty Research Award (2017), Simons Berkeley Research Fellowship (2013), and the IITB President of India Gold Medal (2008). - Faster algorithms for p-norm regression
Abstract:Speaker: Sushant Sachdeva, U. Toronto
Date: 2019-09-03 Time: 16:00:00 (IST) Venue: Bharti Building #501 Linear regression in L_p-norm is a canonical optimization problem that arises in several applications, including sparse recovery, semi-supervised learning, and signal processing. Standard linear regression corresponds to p=2, and p=1 or infinity is equivalent to linear programming.
In this talk, we will discuss new algorithms for obtaining high-accuracy (1+1/poly(n))-approximate solutions to such problems. Our algorithms are based on an ‘iterative refinement’ approach for p-norms, where an algorithm for computing an approximate solution can be converted into a high-accuracy algorithm. Previously, iterative
refinement was known only for L_2-regression (or equivalently, solving linear systems).
Based on this approach, we give several new high-accuracy algorithms, including 1) an algorithm for p-norm regression by solving only m^{1/3} linear systems, 2) a provably convergent and fast iterative reweighted least squares (IRLS) algorithm for p-norm regression, and 3) an algorithm for p-norm flows in undirected unweighted graphs in almost-linear time.
This talk will be based on joint works with Deeksha Adil, Richard Peng, Rasmus Kyng, and Di Wang.
Bio:Sushant Sachdeva is a faculty member at the CS dept. at the University of Toronto. He is interested in Algorithms, and its connections to optimization, machine learning, and statistics. His recent research focus has been the design of fast algorithms for graph problems.
Before joining UofT, he was a research scientist at Google. He completed his postdoc at Yale with Dan Spielman (2016), his PhD from Princeton (2013) under Sanjeev Arora, and his BTech from IIT Bombay (2008). He is the recipient of Google Faculty Research Award (2017), Simons Berkeley Research Fellowship (2013), and the IITB President of India Gold Medal (2008). - An introduction to Text Simplification
Abstract:Speaker: Horacio Saggion
Date: 2019-08-30 Time: 14:15:00 (IST) Venue: SIT Building #001 In this presentation, we aim to provide an extensive overview of automatic text simplification systems proposed so far, the methods they used and discuss the strengths and shortcomings of each of them, providing direct comparison of their outputs. We aim to break some common misconceptions about what text simplification is and what it is not. We believe that understanding of initial motivations, and an in-depth analysis of existing TS methods would help researchers new to ATS propose novel systems, bringing fresh ideas from other related NLP areas. We will describe and explain some of the most influential methods used for automatic simplification of texts so far, with the emphasis on their strengths and weaknesses sometimes showing systems outputs. We will present some of the existing resources for TS for various languages, including parallel manually produced TS corpora, comparable automatically aligned TS corpora, paraphrase- and synonym- resources, TS-specific sentence-alignment tools, and several TS evaluation resources. Finally, we will discuss the existing evaluation methodologies for TS, and necessary conditions for using each of them.
Bio:Horacio Saggion is an Associate Professor at the Department of Information and Communication Technologies, Universitat Pompeu Fabra (UPF), Barcelona. He is the head of the Large Scale Text Understanding Systems Lab, associated to the Natural Language Processing group (TALN) where he works on automatic text summarization, text simplification, information extraction, sentiment analysis and related topics. Horacio obtained his PhD in Computer Science from Universite de Montreal, Canada in 2000. He obtained his BSc in Computer Science from Universidad de Buenos Aires in Argentina, and his MSc in Computer Science from UNICAMP in Brazil. He was the Principal Investigator for UPF in the EU projects Dr Inventor and Able-to-Include and is currently principal investigator of the national project TUNER and the Maria de Maeztu project Mining the Knowledge of Scientific Publications. Horacio has published over 150 works in leading scientific journals, conferences, and books in the field of human language technology. He organized four international workshops in the areas of text summarization and information extraction and was scientific Co-chair of STIL 2009 and scientific Chair of SEPLN 2014. He is a regular programme committee member for international conferences such as ACL, EACL, COLING, EMNLP, IJCNLP, IJCAI and is an active reviewer for international journals in computer science, information processing, and human language technology. Horacio has given courses, tutorials, and invited talks at a number of international events including LREC, ESSLLI, IJCNLP, NLDB, and RuSSIR.
- Mining and Enriching Multilingual Scientific Text Collections
Abstract:Speaker: Prof. Horacio Saggion from Spain
Date: 2019-08-29 Time: 12:00:00 (IST) Venue: SIT Building #001 Scientists worldwide are confronted with an exponential growth in the number of scientific documents being made available, for example: Elsevier publishes over 250K scientific articles per year (or one every two minutes) and has over 7 million publications; MedLine, the most important source in biomedical research, contains 21 million scientific references, and the World Intellectual Patent Organization (WIPO) contains some 70 million records. All this unprecedented volume of information complicates the task of researchers who are faced with the pressure of keeping up-to-date with discoveries in their own disciplines and with the challenge of searching for innovation, new interesting problems to solve, checking already solved problems or hypothesis, or getting information on past and current available methods, solutions or techniques. At the same time and with the rise of open science initiatives and social media, research is more connected and open creating new opportunities but also challenges for the scientific community.
In this scenario of scientific information overload, natural language processing has a key role to play. Over the past few years we have seen a number of tools for the analysis of the structure of scientific documents (e.g. transforming PDF to XML), methods for extracting keywords, or classifying sentences into argumentative categories being developed. However, deep analysis of scientific documents such as: finding key claims, assessing the argumentative quality and strength of the research, or summarizing the key contributions of a piece of work are less common. Besides, most research in scientific text processing is being carried out for the English language, neglecting both the share of scientific information available in other languages and the fact that scientific publications are many times bilingual.
In this talk, I will present work carried out in our laboratory towards the development of a system for “deep” analysis and annotation of scientific text collection. Originally for the English language, it has now being adapted to Spanish. After a brief overview of the system and its main components, I will present the development of a bi-lingual (Spanish and English) fully annotated text resource in the field of natural language processing that we have created with our system together with a faceted-search and visualization system to explore the created resource.
With this scenario in mind I will speculate on the challenges and opportunities that the scientific field brings to our community not only in terms of language but also from the point of view of social media and science education.
Bio:Horacio Saggion is an Associate Professor at the Department of Information and Communication Technologies, Universitat Pompeu Fabra (UPF), Barcelona. He is the head of the Large Scale Text Understanding Systems Lab, associated to the Natural Language Processing group (TALN) where he works on automatic text summarization, text simplification, information extraction, sentiment analysis and related topics. Horacio obtained his PhD in Computer Science from Universite de Montreal, Canada in 2000. He obtained his BSc in Computer Science from Universidad de Buenos Aires in Argentina, and his MSc in Computer Science from UNICAMP in Brazil. He was the Principal Investigator for UPF in the EU projects Dr Inventor and Able-to-Include and is currently principal investigator of the national project TUNER and the Maria de Maeztu project Mining the Knowledge of Scientific Publications. Horacio has published over 150 works in leading scientific journals, conferences, and books in the field of human language technology. He organized four international workshops in the areas of text summarization and information extraction and was scientific Co-chair of STIL 2009 and scientific Chair of SEPLN 2014. He is a regular programme committee member for international conferences such as ACL, EACL, COLING, EMNLP, IJCNLP, IJCAI and is an active reviewer for international journals in computer science, information processing, and human language technology. Horacio has given courses, tutorials, and invited talks at a number of international events including LREC, ESSLLI, IJCNLP, NLDB, and RuSSIR.
- Algorithms for Robust Clustering
Abstract:Speaker: Ravishankar Krishnaswamy, Microsoft Research India Banglore
Date: 2019-08-28 Time: 12:00:00 (IST) Venue: Bharti Building #501 Clustering is one of the fundamental problems in optimization: given a set of points P in a metric space (i.e., any distance function which satisfies the triangle inequality d(u,v) + d(v,w) >= d(u,w)), partition them into k clusters to minimize some objective function (either sum of distances of each point to cluster center, which becomes the k-median problem, or sum of squared distances, which becomes the k-means problem, etc.). While these problems are very classical and widely studied, they all suffer from the drawback of not being "robust" to outliers in the data. That is, one can add a few outlier points to the data-set and completely change the structure of the optimal clustering.
In this talk, we will describe some of our recent works on designing algorithms for "robust clustering", i.e., given a set of points P, and two targets k and m, identify m points to throw out as "outliers", and cluster the remaining points to minimize the clustering objective. We present the first non-trivial approximation algorithms for these problems. We also consider robust versions of another classical clustering problem called correlation clustering. Based on joint works with Deeparnab Chakrabarty, Devvrit, Shi Li, Nived Rajaraman, and Sai Sandeep.
Bio:Ravishankar Krishnaswamy, Microsoft Research India Banglore
- Can machine learning be leveraged to develop combinatorial optimization algorithms?
Abstract:Speaker: Dr. Deepak Ajwani
Date: 2019-08-23 Time: 16:00:00 (IST) Venue: SIT Building #001 Combinatorial optimization problems are ubiquitous across a range of industries. Many of these problems are NP-hard and it is difficult to create practical solutions that can even solve input instances with a few thousand elements, exactly. Considerable time and effort is usually required to develop efficient algorithms for these problems. So, a natural question to ask is, can we automate the process of developing efficient optimization solutions. In the past, this has led to the development of nature-based meta-heuristics, such as Genetic algorithms, Tabu search, Ant colony optimization etc. Recently, machine learning techniques such as reinforcement learning and deep learning have been explored to learn the output directly from the features of the problem instances. In this talk, I will discuss some of the limitations of these approaches and propose a new framework for leveraging machine learning to scale-up optimization algorithms. In contrast to the existing deep-learning based approaches, I will argue that the proposed framework has the potential to automatically develop simple, intuitive optimization algorithms with local features and interpretable models, that can potentially be analyzed for worst-case guarantees on specific instance classes.
Bio:Dr. Deepak Ajwani is an Assistant Professor at the School of Computer Science, University College Dublin. His research is at the confluence of algorithms and data structures (with a focus on scalable graph algorithms), algorithm engineering, combinatorial optimization, semantic analysis and machine learning.
Prior to this appointment, he worked as a research scientist at Nokia Bell Labs. As part of his work at Nokia Bell Labs, he co-lead the design and development of a cognitive computing tool for interpreting, organizing and navigating unstructured content. He received his Ph.D. from Max Planck Institute for Informatics in 2008. He is an alumnus of IIT Delhi, having obtained his B.Tech + M.Tech degrees in Computer Science and Engineering in 2003. - Optimization on manifolds and its application to learning multilingual word embeddings
Abstract:Speaker: Pratik Jawanpuria
Date: 2019-08-19 Time: 11:00:00 (IST) Venue: SIT Building #001 Riemannian geometry is a generalization of the Euclidean geometry. It includes several non-Euclidean spaces such as set of symmetric positive-definite matrices and set of orthogonal matrices, to name a few. Numerous machine learning problems can be cast as an optimization problem on Riemannian manifolds. Examples include principal component analysis, matrix/tensor completion, learning taxonomy, among others.
In this talk, I will discuss our recent work on cross-lingual mapping of word embeddings in supervised setting. Our approach decouples the source-to-target language transformation into various geometrical components, all residing on Riemannian manifolds. Our framework allows seamless generalization from bilingual to multilingual setting as well. Empirical results illustrate the efficacy of the proposed approach in bilingual lexicon induction tasks. The talk includes a brief introduction to optimization on manifolds.
Link to the work: https://www.mitpressjournals.org/doi/full/10.1162/tacl_a_00257
Bio:Pratik Jawanpuria received the B.Tech. and Ph.D. degrees in Computer Science from the Indian Institute of Technology Bombay, in 2009 and 2014, respectively. He was a postdoctoral researcher at the Saarland University, Germany, in 2015. He is currently a senior applied scientist at Microsoft, India. Prior to this, he worked as an applied scientist in the India Machine Learning Team at Amazon. His primary research interests span optimization for machine learning applications.
Homepage: https://pratikjawanpuria.com/ - A quantum wave in computing
Abstract:Speaker: Prof. Umesh Vazirani
Date: 2019-08-19 Time: 15:30:00 (IST) Venue: Bharti Building #501 The field of quantum computation is entering an exciting new period, with small to medium scale machines around the corner. The potential impact of fully scaled quantum computers includes revolutionary new ways of designing new materials, quantum chemistry, optimization and cryptography. Before this can happen great challenges in quantum algorithms, error-correction and quantum science have to be overcome. In this talk I will give a survey of both the potential impact and the challenges.
Bio:Umesh Vazirani is the Strauch Distinguished Professor of Computer Science at University of California, Berkeley, and director of the Berkeley Quantum Information and Computation Center. He has wide interests in the algorithmic foundations of quantum computing, and his 1993 paper with Ethan Bernstein laid the foundations of quantum complexity theory. Besides quantum computation, he has worked on the computational foundations of randomness and on the theory of classical algorithms: he co-invented with Mehta, Saberi and his brother Vijay Vazirani, the bid scaling algorithm for the AdWords auction which is widely used by Internet search companies, and in 2012 he won the Fulkerson Prize with Arora and Rao for an algorithm for graph partitioning. Vazirani is member of the National Academy of Science, and the author of two books: An Introduction to Computational Learning Theory with Michael Kearns (MIT Press) and Algorithms with Sanjoy Dasgupta and Christos Papadimitriou (McGraw Hill).
- Understanding Visual Appearance from Micro Scale to Global Scale
Abstract:Speaker: Kavita Bala & Chair of the Computer Science Department at Cornell University.
Date: 2019-08-13 Time: 16:00:00 (IST) Venue: 4-5pm, Bharti 501,13 Aug 2019 Mixed reality environments require understanding scenes, and then seamlessly blending the rich visual appearance of real and virtual materials to create a compelling user experience. In this talk I will describe our work on modeling and recognizing complex materials, and fine-grained recognition. Using these algorithms as core building blocks we can further understand appearance at a global scale by mining social media to discover patterns across geography and time.
Bio:Kavita Bala is the Chair of the Computer Science Department at Cornell University. She received her S.M. and Ph.D. from the Massachusetts Institute of Technology (MIT), and her B.Tech. from the Indian Institute of Technology (IIT, Bombay). She co-founded GrokStyle (acquired by Facebook), and is a faculty Fellow with the Atkinson Center for a Sustainable Future.
Bala specializes in computer vision and computer graphics, leading research in recognition and visual search; material modeling and acquisition using physics and learning; and material and lighting perception. Bala's work on scalable rendering, Lightcuts, is the core production rendering engine in Autodesk's cloud renderer; and her instance recognition research was the core technology of GrokStyle's visual search engine. Bala has co-authored the graduate-level textbook "Advanced Global Illumination".
Bala has served as the Editor-in-Chief of Transactions on Graphics (TOG), papers Chair of SIGGRAPH Asia 2011, and is onthe Papers Advisory Group for SIGGRAPH. - Deceptive Media Manipulation – An arms race: Learning to defend & deceive
Abstract:Speaker: Dr. Maneesh Kumar Singh & Head R&D, Verisk AI and the Director of the Human and Computation Intelligence Lab
Date: 2019-08-13 Time: 14:15:00 (IST) Venue: CSE Department, Bharti Building - 501 As digital manipulation techniques get better, it is becoming increasingly difficult to distinguish between real and fake data. Being able to model the real world and generate realistic data has many exciting applications. However, it poses a serious challenge when such data is generated by malicious actors with an intent to deceive the user (or the computer system) into believing that the fake data being generated is real. While such attacks on images have garnered much attention, they span across media types including speech & audio, text and, finally, fake news. To counter these attacks, intensive efforts have been undertaken by the ML community to generate effective defense strategies which malicious adversaries, in turn, seek to circumvent. This has led to a virtual arms race.
In this talk, I will provide an overview of the recent efforts in my group at Verisk. We have focused on developing algorithms to defend against expert manipulation of images as well as adversarial attacks that can bypass such defenses. Most of this work is hot off the press and under review at various conferences.
Bio:Dr. Maneesh Kumar Singh is Head R&D, Verisk AI and the Director of the Human and Computation Intelligence Lab. His lab is working on creating collaborative (HMC, MMC) systems for efficient extraction of information & knowledge from unstructured data, capable of closed-loop & explainable reasoning and continuous learning, with applications in vision, NLP, speech and fintech. Verisk Analytics delivers data analytic tools and services for risk assessment, risk forecasting and decision analytics in a variety of sectors including insurance, financial services, energy, government and human resources.
Dr. Singh has over 15 years of customer-focused industrial R&D experience with stints at Siemens CT and SRI International building and delivering image and video analytics technologies in the areas of multi-camera security and surveillance, aerial surveillance, advanced driver assistance and intelligent traffic control; industrial inspection; and, medical image processing and patient diagnostics systems. Dr. Singh received his Ph.D. in Electrical and Computer Engineering from the University of Illinois at in 2003. He has authored over 35 publications (1 best paper award), and has 15+ U.S. and International patents. - AI For Societally Relevant Problems: Influence Maximization in an Uncertain World
Abstract:Speaker: Amulya Yadav, Assistant Professor,Penn State University
Date: 2019-06-03 Time: 10:30:00 (IST) Venue: SIT Building #001 The potential of Artificial Intelligence to tackle challenging problems that afflict society is enormous, particularly in the areas of healthcare, conservation and public safety and security. Many problems in these domains involve harnessing social networks of under-served communities to enable positive change, e.g., using social networks of homeless youth to raise awareness about HIV (and other STDs). Unfortunately, most of these real-world problems are characterized by uncertainties about social network structure and influence models, and previous research in AI fails to sufficiently address these uncertainties, as they make several unrealistic simplifying assumptions for these domains. In this talk, I will describe my research on algorithmic interventions in social networks. In the first part of my talk, I will describe my work on developing new influence maximization algorithms which can handle various uncertainties in social network structure, influence models, etc., that commonly exist in real-world social networks. I will discuss how my algorithms utilize techniques from sequential planning problems and computational game theory to develop new kinds of algorithms in the sub-fields of multi-agent systems and reasoning under uncertainty. In the second part of my talk, I will discuss the real-world deployment of my algorithms to spread awareness about HIV among homeless youth in Los Angeles. This represents one of the first-ever deployments of computer science based influence maximization algorithms in this domain. I will discuss the challenges that I faced, and the lessons that can be gleaned for future deployment of AI systems. Finally, I will also talk about other kinds of societally relevant problems that I have worked on, e.g., raising grievances of low literate farmers to government agencies in emerging market countries, fake news detection, etc. All these problems illustrate the enormous potential of AI in addressing societally relevant problems.
Bio:Amulya Yadav is an Assistant Professor in the College of Information Sciences and Technology at Penn State University. He also has an affiliate faculty appointment with the USC Center for Artificial Intelligence in Society. His research interests include Artificial Intelligence, Multi-Agent Systems, Computational Game-Theory and Applied Machine Learning. His work in the field of Artificial Intelligence for Social Good focuses on developing theoretically grounded approaches to real-world problems that can have an impact in the field. His algorithms have been deployed in the real-world, particularly in the field of public health and wildlife protection. Amulya is a recipient of the AAMAS 2016 Best Student Paper Award, the AAAI 2017 Best Video and Best Student Video Award, the IDEAS 2016 Most Visionary Paper Award, and the AAMAS 2017 Best Paper Award nomination. His work has also been highlighted by Mashable.com as one of 26 incredible innovations that improved the world in 2015.
Amulya holds a Ph.D. in Computer Science from the University of Southern California, and a B. Tech. in Computer Science & Engineering from Indian Institute of Technology (IIT), Patna. - Spatial and spatio-temporal data analytics and their applications in urban cities
Abstract:Speaker: Prof. Cheng Long from NTU
Date: 2019-05-24 Time: 12:00:00 (IST) Venue: Bharti Building #501 It is expected that by 2050, more than 2.5 billion people would reside in cities. While the urbanization has modernized many peoples’ lives, it causes big challenges such as traffic congestion, air pollution, energy consumption, etc. As part of the effort for solving these problems, people have been collecting and analyzing data that is being generated in the urban space, e.g., traffic data, mobility data, POI data, etc., for finding insights into the problems and/or serving citizens better decision making. Since the majority of the data involves the spatial and/or temporal dimensions, techniques for spatial and spatio-temporal data analytics are playing critical roles. In this talk, we will overview this data-driven process, introduce some of its interesting applications, and also present some of our recent work on spatial and spatio-temporal data analytics, including online bipartite matching, co-location pattern mining, and spatiotemporal representation learning, etc.
Bio:LONG Cheng is currently an Assistant Professor at the School of Computer Science and Engineering (SCSE), Nanyang Technological University (NTU). From 2016 to 2018, he worked as a lecturer (Asst Professor) at Queen's University Belfast, UK. He got the PhD degree from the Department of Computer Science and Engineering, The Hong Kong University of Science and Technology (HKUST) in 2015. His research interests are broadly in data management and data mining with his vision to achieve scalable spatial computing, to make sense of urban related data for smarter cities, and to manage and analyze emerging big data such as IoT data for richer knowledge. His research has been recognized with one "Best Research Award" provided by ACM-Hong Kong, one "Fulbright-RGC Research Award" provided by Research Grant Council (Hong Kong), two "PG Paper Contest Awards" provided by IEEE-HK, and one "Overseas Research Award" provided by HKUST. He has served as a Program Committee member/referee for several top data management and data mining conferences/journals (TODS, VLDBJ, TKDE, ICDM, CIKM, etc.). He is member of ACM and IEEE.
- Resource-Aware Session Types for Digital Contracts
Abstract:Speaker: Dr. Ankush Das, CMU
Date: 2019-05-21 Time: 00:00:00 (IST) Venue: CSE Seminar Room, Bharti 501 While there exist several successful techniques for supporting programmers in deriving static resource bounds for sequential code, analyzing the resource usage of message-passing concurrent processes poses additional challenges. To meet these challenges, this talk presents two analyses for statically deriving worst-case bounds on the sequential (work) and parallel (span) complexity of message-passing processes, respectively. The work analysis is based on novel resource-aware session types that describe resource contracts by allowing both messages and processes to carry potential and amortize cost. The span analysis enriches session types with temporal modalities to additionally prescribe the timing of the exchanged messages in a way that is precise yet flexible. The talk finally applies session types to implement digital contracts. Digital contracts are protocols that describe and enforce execution of a contract, often among mutually distrusting parties. Programming digital contracts comes with its unique challenges, which include describing and enforcing protocols of interaction, analyzing resource usage and tracking linear assets. The talk presents the type-theoretic foundations of Nomos, a programming language for digital contracts whose type system addresses the aforementioned domain-specific requirements. To express and enforce protocols, Nomos is based on shared binary session types rooted in linear logic. To control resource usage, Nomos uses resource-aware session types and automatic amortized resource analysis, a type-based technique for inferring resource bounds. To track assets, Nomos employs a linear type system that prevents assets from being duplicated or discarded. The talk concludes with future work directions on designing an efficient implementation for Nomos and making it robust to attacks from malicious entities.
Bio:Ankush Das is a fourth year PhD student at Carnegie Mellon University. He is advised by Prof. Jan Hoffmann. He is broadly interested in programming languages with a specific focus on analysis of resource consumption of programs. In the past, he has worked jointly with Prof. Frank Pfenning and his advisor on designing resource-aware session types for parallel complexity analysis of concurrent programs. Recently, he has been working with Stephanie Balzer, along with Prof. Frank Pfenning and Prof. Jan Hoffmann on designing Nomos, a type safe domain-specific programming language based on resource-aware session types for implementing smart contracts.
Before joining CMU, he worked as a Research Fellow at Microsoft Research, India with Akash Lal where he developed an efficient method to perform precise alias analysis for C and C++ programs for Windows driver modules to automatically infer safe null pointer dereferences.
He completed his undergraduate at IIT Bombay, India in 2015 where he worked with Prof. Supratik Chakraborty and Prof. Akshay S on deciding termination of linear loop programs. - Resource-Aware Session Types for Digital Contracts
Abstract:Speaker: Dr. Ankush Das, CMU
Date: 2019-05-21 Time: 00:00:00 (IST) Venue: IIA- 501 bharti While there exist several successful techniques for supporting programmers in deriving static resource bounds for sequential code, analyzing the resource usage of message-passing concurrent processes poses additional challenges. To meet these challenges, this talk presents two analyses for statically deriving worst-case bounds on the sequential (work) and parallel (span) complexity of message-passing processes, respectively. The work analysis is based on novel resource-aware session types that describe resource contracts by allowing both messages and processes to carry potential and amortize cost. The span analysis enriches session types with temporal modalities to additionally prescribe the timing of the exchanged messages in a way that is precise yet flexible. The talk finally applies session types to implement digital contracts. Digital contracts are protocols that describe and enforce execution of a contract, often among mutually distrusting parties. Programming digital contracts comes with its unique challenges, which include describing and enforcing protocols of interaction, analyzing resource usage and tracking linear assets. The talk presents the type-theoretic foundations of Nomos, a programming language for digital contracts whose type system addresses the aforementioned domain-specific requirements. To express and enforce protocols, Nomos is based on shared binary session types rooted in linear logic. To control resource usage, Nomos uses resource-aware session types and automatic amortized resource analysis, a type-based technique for inferring resource bounds. To track assets, Nomos employs a linear type system that prevents assets from being duplicated or discarded. The talk concludes with future work directions on designing an efficient implementation for Nomos and making it robust to attacks from malicious entities.
Bio:Bio :-
Ankush Das is a fourth year PhD student at Carnegie Mellon University. He is advised by Prof. Jan Hoffmann. He is broadly interested in programming languages with a specific focus on analysis of resource consumption of programs. In the past, he has worked jointly with Prof. Frank Pfenning and his advisor on designing resource-aware session types for parallel complexity analysis of concurrent programs. Recently, he has been working with Stephanie Balzer, along with Prof. Frank Pfenning and Prof. Jan Hoffmann on designing Nomos, a type safe domain-specific programming language based on resource-aware session types for implementing smart contracts.
Before joining CMU, he worked as a Research Fellow at Microsoft Research, India with Akash Lal where he developed an efficient method to perform precise alias analysis for C and C++ programs for Windows driver modules to automatically infer safe null pointer dereferences.
He completed his undergraduate at IIT Bombay, India in 2015 where he worked with Prof. Supratik Chakraborty and Prof. Akshay S on deciding termination of linear loop programs. - Resource-Aware Session Types for Digital Contracts
Abstract:Speaker: Ankush Das, CMU
Date: 2019-05-21 Time: 12:15:00 (IST) Venue: Bharti Building #501 While there exist several successful techniques for supporting programmers in deriving static resource bounds for sequential code, analyzing the resource usage of message-passing concurrent processes poses additional challenges. To meet these challenges, this talk presents two analyses for statically deriving worst-case bounds on the sequential (work) and parallel (span) complexity of message-passing processes, respectively. The work analysis is based on novel resource-aware session types that describe resource contracts by allowing both messages and processes to carry potential and amortize cost. The span analysis enriches session types with temporal modalities to additionally prescribe the timing of the exchanged messages in a way that is precise yet flexible. The talk finally applies session types to implement digital contracts. Digital contracts are protocols that describe and enforce execution of a contract, often among mutually distrusting parties. Programming digital contracts comes with its unique challenges, which include describing and enforcing protocols of interaction, analyzing resource usage and tracking linear assets. The talk presents the type-theoretic foundations of Nomos, a programming language for digital contracts whose type system addresses the aforementioned domain-specific requirements. To express and enforce protocols, Nomos is based on shared binary session types rooted in linear logic. To control resource usage, Nomos uses resource-aware session types and automatic amortized resource analysis, a type-based technique for inferring resource bounds. To track assets, Nomos employs a linear type system that prevents assets from being duplicated or discarded. The talk concludes with future work directions on designing an efficient implementation for Nomos and making it robust to attacks from malicious entities.
Bio:Ankush Das is a fourth year PhD student at Carnegie Mellon University. He is advised by Prof. Jan Hoffmann. He is broadly interested in programming languages with a specific focus on analysis of resource consumption of programs. In the past, he has worked jointly with Prof. Frank Pfenning and his advisor on designing resource-aware session types for parallel complexity analysis of concurrent programs. Recently, he has been working with Stephanie Balzer, along with Prof. Frank Pfenning and Prof. Jan Hoffmann on designing Nomos, a type safe domain-specific programming language based on resource-aware session types for implementing smart contracts.
Before joining CMU, he worked as a Research Fellow at Microsoft Research, India with Akash Lal where he developed an efficient method to perform precise alias analysis for C and C++ programs for Windows driver modules to automatically infer safe null pointer dereferences.
He completed his undergraduate at IIT Bombay, India in 2015 where he worked with Prof. Supratik Chakraborty and Prof. Akshay S on deciding termination of linear loop programs. - Debugging and Optimizing Code in a world of Software Frameworks
Abstract:Speaker: Dr Abhilash Jindal
Date: 2019-05-20 Time: 00:00:00 (IST) Venue: IIA-501 Bharti Modern programmers stand on "the shoulders of giant" software frameworks. The programmers are able to build complicated applications fairly quickly by just stringing together many intricate framework APIs. But, due to the limited understanding of underlying frameworks, programmers often make mistakes in using these APIs. Moreover, this limited understanding has made debugging software systems and improving their performance significantly harder.
In this talk, we first take a close look at wakelock APIs, a set of APIs for explicitly managing the power states of hardware components, that was introduced by the Android framework. I will present the complexity of correctly using wakelock APIs and present a taxonomy of most prevalent energy bugs in the Android ecosystem— sleep disorder bugs which arise from incorrect usage of wakelock APIs. I will then present KLOCK that makes use of a set of static analyses to systematically identify sleep-induced time bugs, a subclass of sleep disorder bugs.
In the second part, we study performance optimization in programming high-level framework APIs. The state of the art for finding general optimizations is through profiling that can help developers identify hotspots, i.e, method calls that contribute to a large portion of the resource consumption in their source code. In Android apps, the hotspots identified by the profiler are almost always framework API calls. Since developers did not author framework APIs, even after they are presented with the API hotspots, they typically do not have any guidance on how to proceed with the remaining optimization process: (1) Is there a more efficient way of using framework APIs to implement the same app task? (2) How to come up with more efficient implementation? To help developers tackle these challenges, we developed a new optimization methodology called differential profiling that automatically uncovers more efficient API usage by leveraging existing implementations of similar apps which are bountiful in the app marketplace.
Bio:Abhilash Jindal received B.Tech. in Electrical Engineering from IIT Kanpur. He received his Ph.D. in Computer Science from Purdue University where he researched software-based approaches for extending mobile battery life. He has published papers in top system conferences including OSDI, ATC, Eurosys, Mobisys, Sigmetrics, and Mobicom. Currently, he is commercializing his research work serving as the CTO of Mobile Enerlytics, a silicon valley startup. His research interests include mobile systems, software engineering, and operating systems.
- Computing in the age of low memory
Abstract:Speaker: Dr. Amitabh Trehan
Date: 2019-04-26 Time: 12:00:00 (IST) Venue: Bharti Building #501 We look at the problem of computing in the setting of large networks of low memory! - as may possibly be in setting such as the Internet of Things. We discover a connection between distributed computing and streaming. We formalise this by introducing the Compact Local Streaming model - nodes in an arbitrary network have low memory (at most O(log^2 n)) even though they support O(n) neighbours and computation happens by `local streaming' at nodes.
We look at some simple variants of the model and then propose the first self-healing compact routing protocol. Our scheme, based on Thorup and Zwick [SPAA 2001] tree routing scheme and the ForgivingTree [PODC' 2008] self-healing scheme is a compact scheme (using at most O(log^2 n) memory per node) that functions despite node failures with only a small additional overhead.
This research was supported by the EPSRC first grant COSHER: Compact Self-Healing Routing
References:
[1] Armando Castañeda [1], Jonas Lefèvre [2], Amitabh Trehan[3],
Some Problems in Compact Message Passing -
https://arxiv.org/abs/1803.03042,
[2] Armando Castañeda [4], Danny Dolev [5], Amitabh Trehan: Compact routing messages in self-healing trees. Theoretical Computer Sci [6]ence, 2018,
https://www.sciencedirect.com/science/article/pii/S0304397516306818?via%3Dihub [7]
Bio:Dr. Amitabh Trehan is an assistant professor of Computer Science at Loughborough University, specailising in algorithms, in partcular, distributed and reliable (self-healing) algorithms. He received his PhD from University of New Mexico, USA, followed by postdocs in Canada (Victoria) and Israel(Technion, Hebrew University). Before going to the US, he did a BSc from Punjab University and Master's from IGNOU and IIT Delhi.
- Opinion Dynamics in Social Networks: Modeling and Inference
Abstract:Speaker: Dr. Abir De
Date: 2019-04-25 Time: 12:00:00 (IST) Venue: Bharti Building #501 Social media and social networking sites have become a global pinboard for exposition and discussion of news, topics, and ideas, where the users often update their opinions about a particular topic by learning from the opinions shared by their friends. Understanding such networked opinion formation process has a large spectrum of applications, for example, poll prediction, brand estimation, etc. Beyond prediction, news and information are often curated by editorial boards, professional journalists who post feeds on the users' walls in order to steer their opinion to a desired state. Therefore, the underlying opinion dynamical process also involves the influence of external sources--- which are difficult to identify. In this talk, I shall present these challenging tasks related to opinion dynamics from all these perspectives--- (a) learning a data-driven model of spontaneous opinion exchange; and (b) demarcating endogenous and exogenous opinion diffusion process in social networks. Finally, I will present some interesting directions for future research.
Bio:Dr. Abir De , Professor
- Designing intelligent machines via reactive systems
Abstract:Speaker: Suguman Bansal, Rice University, Houston
Date: 2019-04-24 Time: 12:00:00 (IST) Venue: Bharti Building #501 Intelligent machines, such as IoT devices, are fundamentally reactive systems that interact with the outer physical environment to achieve a goal. Asynchronous interactions are ubiquitous in reactive systems and complicate the design and programming of reactive systems. Automatic construction of asynchronous reactive programs from specifications ("synthesis") could ease the difficulty, but known methods are complex, and intractable in practice.
In this talk, I will formulate a direct, exponentially more compact automaton construction for the reduction of asynchronous to synchronous synthesis. Experiments with a prototype implementation of the new method demonstrate practical feasibility. Furthermore, it is shown that for several useful classes of temporal properties, automaton-based methods can be avoided altogether and replaced with simpler Boolean constraint solving.
Bio:Suguman Bansal is a PhD candidate in the Department of Computer Science at Rice University. Her research interest spans formal methods, specifically the application of quantitative verification and reactive synthesis. She was a visiting graduate student at the Simons Institute Spring Program on "Real-Time Decision Making" at the University of California, Berkeley in 2018, and has spent the summers of 2017 and 2018 interning at Nokia Bell Labs. She is the recipient of the Andrew Ladd Fellowship, and a Gold Medal at the Association for Computing Machinery (ACM) Student Research Competition at POPL.
- Designing intelligent machines via reactive systems
Abstract:Speaker: Suguman Bansal, Rice University, Houston
Date: 2019-04-24 Time: 12:00:00 (IST) Venue: Bharti Building #501 Intelligent machines, such as IoT devices, are fundamentally reactive systems that interact with the outer physical environment to achieve a goal. Asynchronous interactions are ubiquitous in reactive systems and complicate the design and programming of reactive systems. Automatic construction of asynchronous reactive programs from specifications ("synthesis") could ease the difficulty, but known methods are complex, and intractable in practice.
In this talk, I will formulate a direct, exponentially more compact automaton construction for the reduction of asynchronous to synchronous synthesis. Experiments with a prototype implementation of the new method demonstrate practical feasibility. Furthermore, it is shown that for several useful classes of temporal properties, automaton-based methods can be avoided altogether and replaced with simpler Boolean constraint solving.
Bio:Suguman Bansal is a PhD candidate in the Department of Computer Science at Rice University. Her research interest spans formal methods, specifically the application of quantitative verification and reactive synthesis. She was a visiting graduate student at the
Simons Institute Spring Program on "Real-Time Decision Making" at the University of California, Berkeley in 2018, and has spent the summers of 2017 and 2018 interning at Nokia Bell Labs. She is the recipient of the Andrew Ladd Fellowship, and a Gold Medal at the Association for Computing Machinery (ACM) Student Research Competition at POPL. - Static Analysis and Automated Verification for Replicated Systems
Abstract:Speaker: Dr Kartik Nagar , Purdue University
Date: 2019-04-15 Time: 12:00:00 (IST) Venue: Bharti Building #501 Replicated systems provide a number of desirable properties such as scalability, always-on availability, fault tolerance and low latency in a geo-distributed setting. However, writing correct applications for such systems is inherently hard due to their non-sequential nature and unbounded concurrency. To make such systems easier to program, a number of so-called weakly consistent replicated data stores have emerged in the last few years, providing a tradeoff between consistency and performance. It is still a monumental task for programmers to find the weakest consistency level under which their applications can run correctly. In this talk, I will present some of my recent work which addresses this problem by proposing static analysis techniques to reason about the correctness of programs in replicated systems. Notably, the proposed techniques are parametric in the weak consistency model, and hence given an application and a specification of the replicated store, they can automatically find either a correctness bug if it exists, or verify that the application will run correctly. Different techniques are proposed targeting different correctness criteria used by programmers, such as preservation of state-based invariants, serializability for database applications and convergence for high-level data types.
Bio:Kartik Nagar is a postdoctoral research associate at Purdue University working with Prof. Suresh Jagannathan His research interests are in Program Analysis and Automated Verification. He completed his PhD from Indian Institute of Science under the guidance of Prof. Y N Srikant, and his PhD thesis was on static timing analysis of programs for Real-time systems.
- On optimal ε-nets for set systems
Abstract:Speaker: Dr. Kunal Dutta
Date: 2019-04-03 Time: 12:00:00 (IST) Venue: Bharti Building #501 This talk shall have three main parts. First, I shall give a brief overview of my research areas and contribution. Next, I shall present some recent results on optimal bounds ε-nets for set systems in terms of their shallow cell complexity, using the shallow packing lemma of D.-Ezra-Ghosh and Mustafa. Finally, I shall conclude with some slides on my research and teaching interests.
{em ε-nets}, introduced by Haussler and Welzl (1987), are compact representations of set systems. In the three decades since their introduction, they have found applications in several areas, ranging from statistical inference and machine learning, to algorithms, databases and combinatorial optimization. Finding optimal-sized ε-nets for various set systems, therefore, has been one of the most widely investigated areas in Computational Geometry. A culminating point in this direction, are the recent results of several authors, bounding the size of ε-nets for set systems in terms of their shallow cell complexity. In a related direction, the celebrated {em packing lemma} of Haussler (1991) states that given a set system of bounded VC dimension, any sub-collection of sets with large pairwise symmetric difference, cannot have too many sets. Recently it was generalised by D.-Ezra-Ghosh (2015) and by Mustafa (2016) to the {em shallow packing lemma}.
I shall present a very short proof of optimal ε-nets for set systems in terms of their shallow cell complexity, using the shallow packing lemma. This implies all known cases of results on unweighted nets studied for the past 30 years, starting from the result of Matouv{s}ek, Seidel and Welzl (1990) to that of Clarkson and Varadarajan (2007) to that of Varadarajan (2010) and Chan, Grant, K"{o}nemann and Sharpe (2012) for the unweighted case, as well as the technical and intricate paper of Aronov, Ezra and Sharir (2010). Based on joint works with Arijit Ghosh, Nabil Mustafa and Esther Ezra.
Bio:-
- "Dimensional reduction and molecular signature extraction from large-scale genomics datasets"
Abstract:Speaker: Dr. Carl HERRMANN, University Clinics Heidelberg
Date: 2019-03-29 Time: 16:00:00 (IST) Venue: Bharti Building #501 Recent advances in high-throughput sequencing have led to a spectacular increase in the amount of molecular sequencing data available. These techniques allow to track genome-wide gene expression (RNA-seq), as well as sequence features (whole genome sequencing) or epigenetic aspects such as histone modifications or DNA methylation. Together with phenotypic or clinical information, these data represent a comprehensive picture of the cellular processes, for example in disease. However, there are numerous computational challenges, related to dimension of these datasets, and to the diversity of the molecular readouts. Hence, there is a challenge of (1) dimensionality reduction in order to be able to capture essential aspects and visualize those; and (2) data integration in order to combine these distinct views into a unified representation. The goal is to extract lower-dimensional ,molecular signatures” which might correspond for example to specific subtypes in a disease cohort.
Bio:Prof. University Clinics Heidelberg
- A Talk by
Abstract:Speaker: Dr. Pushpendra Singh
Date: 2019-03-25 Time: 12:00:00 (IST) Venue: Bharti Building #501 ith the increasing use of mobile phones and applications by every section of society, the mobile phone has emerged as a truly ubiquitous computing platform. Similarly, with the emerging IoT devices, we are entering the era of ubiquitous computing and sensing. Today, the data collected/generated from the sensors, IoT devices, and mobile phones is being used in different domains e.g. energy, environment, transport, and health. To collect, manage, and analyze this data, we need middleware solutions. Similarly, mobile applications, supported by middleware, have a huge potential in filling the technology gap that is prevalent in rural India especially in the areas of public health and education. I will provide a brief overview of Middleware research in my group in the area of Energy. Then, I will discuss one project from middleware and mobile crowdsensing aspect and another project, a mobile phone-based platform, for training community health workers.
Our mobile crowdsensing work provides deep insight into the impact of battery consumption pattern on task allocation algorithms. We show that different battery usage patterns affect the performance of the task allocation algorithms. Our work provides an important insight into factors affecting the performance of allocation algorithms and advocates incorporating battery usage patterns for future development of these algorithms. None of the existing algorithms currently incorporate battery usage patterns. I will also discuss future research directions in this area.
Inadequate training of community health workers (CHW) has known to affect the outcomes of health interventions. Given the increasingly important role of CHWs in rural areas, there is a need to provide a low-cost and scalable technology platform for training them while compensating for lack of instructors/trainers. Our platform - Sangoshthi - provides a technical platform to support large-scale training of CHWS using a combination of feature phones and a single smartphone. The system has been deployed and is being used for training CHWs. The training has also resulted in statistically significant gains of knowledge. This research is now funded by the Bill & Melinda Gates Foundation. I will present the system and also explain the future research plans in this area. I will conclude with other technology-led initiatives in the area of public health.
Bio:Pushpendra Singh is an associate professor in the dept. of Computer Science and Engineering and the dept. of Human-Centered Design. His research interests are in the areas of mobile middleware systems and HCI for development.
- Opportunities and Challenges in Middleware Systems and Mobile Applications
Abstract:Speaker: Dr. Pushpendra Singh
Date: 2019-03-25 Time: 12:00:00 (IST) Venue: Bharti Building #501 With the increasing use of mobile phones and applications by every section of society, the mobile phone has emerged as a truly ubiquitous computing platform. Similarly, with the emerging IoT devices, we are entering the era of ubiquitous computing and sensing. Today, the data collected/generated from the sensors, IoT devices, and mobile phones is being used in different domains e.g. energy, environment, transport, and health. To collect, manage, and analyze this data, we need middleware solutions. Similarly, mobile applications, supported by middleware, have a huge potential in filling the technology gap that is prevalent in rural India especially in the areas of public health and education. I will provide a brief overview of Middleware research in my group in the area of Energy. Then, I will discuss one project from middleware and mobile crowdsensing aspect and another project, a mobile phone-based platform, for training community health workers.
Our mobile crowdsensing work provides deep insight into the impact of battery consumption pattern on task allocation algorithms. We show that different battery usage patterns affect the performance of the task allocation algorithms. Our work provides an important insight into factors affecting the performance of allocation algorithms and advocates incorporating battery usage patterns for future development of these algorithms. None of the existing algorithms currently incorporate battery usage patterns. I will also discuss future research directions in this area.
Inadequate training of community health workers (CHW) has known to affect the outcomes of health interventions. Given the increasingly important role of CHWs in rural areas, there is a need to provide a low-cost and scalable technology platform for training them while compensating for lack of instructors/trainers. Our platform - Sangoshthi - provides a technical platform to support large-scale training of CHWS using a combination of feature phones and a single smartphone. The system has been deployed and is being used for training CHWs. The training has also resulted in statistically significant gains of knowledge. This research is now funded by the Bill & Melinda Gates Foundation. I will present the system and also explain the future research plans in this area. I will conclude with other technology-led initiatives in the area of public health.
Bio:Pushpendra Singh is an associate professor in the dept. of Computer Science and Engineering and the dept. of Human-Centered Design. His research interests are in the areas of mobile middleware systems and HCI for development.
- "Congestion Control for Networks with Link and Traffic Variability"
Abstract:Speaker: Prateesh Goyal
Date: 2019-03-20 Time: 15:00:00 (IST) Venue: SIT Building #001 Achieving high throughput and low delay has been a primary motivation for congestion control researchers for decades. Unlike wired links, wireless links exhibit variations in available capacity making this task particularly challenging. This talk will present Accel-Brake Control (ABC), a simple and deployable explicit congestion control protocol for network paths with time-varying wireless links. ABC routers mark each packet with an "accelerate" or "brake", which causes senders to slightly increase or decrease their congestion windows. Routers use this feedback to quickly guide senders towards a desired target rate.
The talk will also describe Nimbus, a congestion control system for wide-area networks with traffic variability. Nimbus enables a sender to reduce queuing delays without compromising fairness with cross traffic. Nimbus employs a novel technique to detect whether the cross traffic competing with a flow is elastic or not, and uses the elasticity detector to improve congestion control by explicitly switching operating modes.
Bio:Prateesh Goyal is a PhD student in the Networks and Mobile Systems group at MIT CSAIL, advised by Hari Balakrishnan and Mohammad Alizadeh. He is broadly interested in improving quality of service for Internet users. He and his colleagues created the ABC and Nimbus protocols for better congestion control over wireless (Wi-Fi and Cellular) and wide-area networks respectively. He has also worked adaptive bit rate algorithms for video streaming (Minerva), implementing congestion control outside the data plane (CCP), and monitoring network performance (Marple). He received the SIGCOMM Best Paper Award, and the Institute Silver Medal and Shankar Dayal Gold Medal at IIT Bombay.
- Challenges and Opportunities in Motion and Eye Tracking
Abstract:Speaker: Prof. Eakta Jain, University of Florida
Date: 2019-03-19 Time: 12:00:00 (IST) Venue: Bharti Building #501 We are on the cusp of ubiquitous human tracking. With Internet of Things and pervasive sensing, we can expect that our motion will be tracked, our faces will be recorded and interpreted, and our bio-measurements will be routinely monitored. My research considers two aspects of ubiquitous human tracking: motion tracking and eye tracking. In this talk, I will discuss one project from each aspect.
First: While large databases of adult motion are commoditized today, the same type of data is hard to collect for young children. However this data is extremely valuable to develop motion models for motion tracking and synthetic motion generation. Given the difficulty of motion capturing children, I will discuss our research organized around two questions: (a) can people perceive if a motion came from a child or an adult in the absence of appearance information? (b) can we algorithmically transform adult motion capture data to look child-like?
Second: Large databases of eye-tracking data have been collected and used to generate computational models of visual saliency. While streaming data from bio-sensors such as EDA, HR, BVP have been used for performance analysis and affective analysis, eye-tracking data has been largely ignored for these applications. As eye-tracking becomes a built-in service for virtual and augmented reality headsets, we ask the question: can we extract an index of affect from eye-tracking data? I will discuss our research aimed at leveraging the pupil as an index of user engagement.
Bio:Eakta Jain is an Assistant Professor of Computer and Information Science and Engineering at the University of Florida. She received her PhD and MS degrees in Robotics from Carnegie Mellon University and her B.Tech. degree from IIT Kanpur. She has worked in industrial research at Texas Instruments R&D labs, Disney Research Pittsburgh, and the Walt Disney Animation Studios. Her research group at the University of Florida is funded through faculty research awards from Facebook/Oculus and Google/YouTube, federal funding from the National Science Foundation, and state funding from the Florida Department of Transportation.
- Policy Enforcement Across Web Applications
Abstract:Speaker: Abhishek Bichhawa,Carnegie Mellon University, USA.
Date: 2019-03-14 Time: 12:00:00 (IST) Venue: Bharti Building #501 Web applications routinely access sensitive and confidential data through remote APIs. The privacy of such data is governed by complex policies implemented as inline checks across application code and database queries. However, the complexity of the policies and the code often results in missed policy checks that cause unauthorized information leaks. We present Estrela, a framework to build Web applications, that allows specification of rich and expressive policies separately from the code, and enforces it based on what and how the data is being accessed on the server-side. On the client-side, we build an information flow tracking policy specification and enforcement mechanism in Web browsers for fine-grained policy enforcement specified by developers.
Bio:Abhishek Bichhawat is a postdoctoral fellow in the School of Computer Science at Carnegie Mellon University, USA. His research focuses on language-based security, program analysis and verification, and building secure-by-design systems. He received a PhD from Saarland University where he was affiliated with CISPA and the International Max Planck Research School for Computer Science.
- Amplification Theorems for Differentially Private Machine Learning
Abstract:Speaker: Kunal Talwar (Google)
Date: 2019-03-13 Time: 12:00:00 (IST) Venue: Bharti Building #501 A rigorous foundational approach to private data analysis has emerged in theoretical computer science in the last decade, with differential privacy and its close variants playing a central role. Recent works have shown that complex machine learning models can be trained with little accuracy loss, while giving strong differentially privacy guarantees. The analyses of these algorithms rely on a class of results known as privacy amplification theorems. In this talk, I will describe two recent results of this kind.
(Joint works with Ulfar Erlingsson, Vitaly Feldman, Ilya Mironov, Ananth Raghunathan and Abhradeep Thakurta)
Bio:Kunal Talwar (Google)
- Relaxed Memory Concurrency and Compiler Correctness
Abstract:Speaker: Soham Chakraborty, MPI-SWS
Date: 2019-03-07 Time: 12:00:00 (IST) Venue: Bharti Building #501 In most analysis/verification techniques, the behavior of concurrent programs is understood in terms of thread interleavings, a model formally known as sequential consistency (SC). For performance reasons, modern hardware implementations allow some executions which cannot be explained by thread interleavings. Similarly, standard compiler optimizations can affect the outcome of a concurrent programs in ways that cannot be understood by SC. Therefore, to program concurrent systems correctly and effectively, it is important to come up with a programming language semantics that accounts for all the non-SC behaviors that hardware and compilers produce. Defining such a "relaxed" memory model is challenging due to conflicting requirements: the model should not only enable efficient compilation, but also provide programmability guarantees.
In this talk I will introduce relaxed memory consistency and present our work on formally defining a good concurrency model for C/C++. As an application of our model, I will also present a translation validator for the optimizations of LLVM, a state-of-the-art C/C++ compiler. The validator has exposed bugs in LLVM concerning the compilation of concurrent C/C++ programs and has revealed interesting differences between the C/C++ and LLVM concurrency semantics.
Bio:Soham Chakraborty is a PhD candidate in Max Planck Institute for Software Systems (MPI-SWS), Germany. His main research interests are relaxed memory concurrency and compiler correctness. He received the BE in Computer Science and Engineering from Vidyasagar University and the MS in Computer Science and Engineering from IIT Kharagpur. He has worked in various industrial research and development positions for 5 years before joining PhD.
- Common Sense Knowledge for Natural Language Understanding
Abstract:Speaker: Dr Ashutosh Modi, Disney Research LA
Date: 2019-03-06 Time: 12:00:00 (IST) Venue: Bharti Building #501 Humans have always envisioned creating machines that could assist in daily chores. With this vision in mind, the field of Artificial Intelligence (AI) was introduced in 1956 by John McCarthy and colleagues. Given the fact that natural mode of communication for humans is the language they speak/write (a.k.a. natural language), it is desirable to have machines which can communicate using the same. However, developing computer models of natural language is far from trivial. Language comes with its own complexities and inherent ambiguities. In this talk, I would show how one could leverage common sense knowledge about the world to make the task of natural language understanding (NLU) easier.
I would present research on modeling common sense knowledge and its integration in NLU systems. I would present empirical results showing the contribution of common sense knowledge to language comprehension in humans. Later part of the talk would show how social common sense in the form of affects (e.g. emotions) can be useful for NLU. I would further show affective information could be integrated into a conversational system in order to make it more engaging for humans.
Bio:Dr. Ashutosh Modi is a postdoctoral researcher at Disney Research Los Angeles. He mainly researches on developing models of human languages in order to bridge the communication between humans and machines. In particular, he focuses on affective computing, conversational systems and Natural Language Understanding (NLU). He received his doctorate degree from Saarland University, Germany for a thesis on "Modeling Common Sense Knowledge via Scripts". His doctoral research focused on modeling common sense knowledge about prototypical activities like "making a coffee", "going to a restaurant", etc. In the past, he has worked at Siemens Research where he worked on bio-medical question answering system. Prior to that, he worked at Symantec Corporation in the area of email security with the focus on the application of machine learning techniques. Ashutosh has obtained a Masters Degree from IIT, Delhi (Indian Institute of Technology, Delhi) with specialization in digital communication and signal processing.
- Efficient Optimization for Machine Learning
Abstract:Speaker: Elad Hazan, Princeton
Date: 2019-02-27 Time: 16:00:00 (IST) Venue: Bharti Building #501 In this talk we will describe recent advances in optimization that gave rise to state-of-the-art algorithms in machine learning. While stochastic gradient descent is the workhorse behind the recent deep learning revolution, improving its performance both theoretically and in practice has proven challenging. We will explore recent innovations in stochastic second order methods and adaptive regularization that yield some of the fastest algorithms known.
Bio:Professor, CSE department, Princeton University
- Optimization-Based Model Validation – a Mathematical Technology of High Impact
Abstract:Speaker: Prof. Dr. Ekaterina Kostina
Date: 2019-02-25 Time: 16:00:00 (IST) Venue: Bharti Building #501 Validated dependable mathematical models are of ever growing importance in science and engineering, and today also driven by a serious demand of the industry. Apart from providing scientific insight into complex nonlinear processes, mathematical models are fundamental for process simulation, optimization and control. In this context the new paradigmatic concept of “digital twins” may be worth mentioning, which is becoming more and more attractive among researchers from different industrial companies, and is advocated by NASA, GE, Siemens and others. Its idea is that every new system, product or a physical or economical process should be accompanied by a "digital twin", which comprises an ensemble of mathematical models and algorithms. This collection of models, analysis and solution algorithms – a virtual avatar or virtual double – would ideally accompany the process from the "origin" through its life time, "grow" together with this process, and be used among others to analyze data, predict malfunctioning and perform optimal operation. However, the application of mathematical models for the simulation, and even more for the optimization and control of complex engineering processes requires their thorough validation and calibration by parameter and state estimation based on process data, which should preferably be obtained by optimized experiments, and optimized measurement designs. For the latter, very efficient numerical methods for the complex non-standard optimal control problem of optimal experimental design for dynamic processes were developed, which have proven to be a powerful mathematical instrument/technology of high economic impact in industrial practice.
The talk will address new developments in optimization methods for validation and calibration of models, in particular the design of robust optimal experiments based on a second order sensitivity analysis of parameter estimates and design of optimal experiments for PDE models.
Bio:Prof. Dr. Ekaterina Kostina , Institute for Applied Mathematics, Heidelberg University
- "Towards an Intelligence Architecture for Human-Robot Teaming"
Abstract:Speaker: Rohan Paul
Date: 2019-02-20 Time: 12:00:00 (IST) Venue: Bharti Building #501 Advances in autonomy are enabling intelligent robotic systems to enter human-centric environments like factories, homes and workplaces. To be effective as a teammate, we expect robots to accomplish more than performing simplistic repetitive tasks; they must perceive, reason, perform cognitive tasks in a human-like way. A system's ability to act intelligently is fundamentally tied to its understanding of the world around it. In this talk, I will present recent work on constructing an intelligence architecture centered around a semantic world model that can be learned from observations, background knowledge and human guidance enabling the robot to follow complex natural language commands. I will discuss how an intelligent agent can interact with the environment to fill gaps in its knowledge about the world. Finally, I will draw connections with prior work in embodied assistive devices for persons with blindness.
Bio:Rohan is a Postdoctoral Associate at the Computer Science and AI Laboratory at the MIT. His research focuses on grounding natural language instructions and concept acquisition from language and vision; aimed towards capable service robots that interact seamlessly with humans. He pursued a D.Phil. degree at the University of Oxford as a Rhodes Scholar from India. His doctoral research contributed Bayesian models for large-scale semantic mapping and navigation. Rohan obtained a Dual Degree in Computer Science and Engineering and served as a Postdoctoral Fellow at IIT Delhi (2012-2015) contributing to the development of affordable assistive devices for persons with visual impairment, particular aimed towards mobility and education. Rohan's research has received Best Paper Awards at RSS 2016, IROS 2013, TRANSED 2012/2010 and ICRA 2010 (Best Vision Paper Finalist). He is a recipient of the Bracken Bequest Graduate Research Prize at Oxford University, INAE National Award, BOSS and FITT Award at IIT Delhi. He was a member of the team that received the National Award for Outstanding Applied Research Improving the Life of Persons with Disabilities conferred by the Govt. of India and was named one of "35 Global Innovators Under 35" by MIT Technology Review in 2016.
- Cyber-Security Engineering via the Lens of Decision Science - The Case of Social-Networked Distributed Systems
Abstract:Speaker: Ranjan Pal
Date: 2019-02-19 Time: 12:00:00 (IST) Venue: Bharti Building #501 It is widely known in the security research community that the science and engineering of cyber-security is solely not a single dimensional technical exercise, specifically for settings with distributed networked entities. Strengthening cyber-security is a challenging multi-stakeholder, multi-dimensional problem, where the "optimal" solution approach to modern security systems design requires technology to be often aptly complemented with ideas from decision-scientific disciplines such as game theory, AI/learning, and computational economics (both mathematical and behavioral), to name a few. In this talk, I will touch upon the idea of cyber-insurance driven security engineering - an inter-disciplinary decision-scientific engineering approach, to improve networked distributed systems security. In this context, i will focus on the design of (a) an AI-driven computationally efficient cyber-risk estimation methodology, (b) market efficient insurance solutions that improve network security with the view to mutually satisfying multiple conflicting stakeholders, and (c) a Stackelberg game driven smart security pricing framework for networks that computationally efficiently alleviates drawbacks of aforementioned market solutions to result in a_win-win_ state for all stakeholders.
Bio:Ranjan Pal is a visiting Fellow in the department of Computer Science and Technology at the University of Cambridge, hosted by Prof. Jon Crowcroft, and a member of the Cambridge Trust and Technology Initiative. Prior to this, he was a Research Scientist at the University of Southern California (USC), where he was affiliated with both the Electrical Engineering and Computer Science departments. His primary research interests lie in the applications of decision science tools to problems in cyber-security, privacy, and resource management in distributed systems. Ranjan received his PhD in Computer Science from USC in 2014, and was the recipient of the Provost Fellowship throughout his PhD studies. During his graduate studies, Ranjan held visiting student researcher positions at Princeton University, Deutsch Telekom Research Laboratories (T-Labs) Berlin, and Aalborg University. Ranjan is an active industry consultant in cyber-security, and is currently a technical advisor to QxBranch. He is a member of the IEEE, ACM, American Mathematical Society (AMS), and the Game Theory Society.
- Fault-Tolerant Structures in Directed Graphs
Abstract:Speaker: Keerti Choudhary (Weizmann)
Date: 2019-02-05 Time: 12:00:00 (IST) Venue: Bharti Building #501 In this talk, we look at the problem of single-source-reachability (SSR) in presence of failures of vertices and edges. In the static setting, SSR can be solved in O(m+n) time, for any directed graph G on n vertices and m edges. To model the scenario of a faulty network, we associate a parameter k with the network such that there are at most k vertices (or edges) that are failed at any stage in the network. The goal is to preprocess the graph G in polynomial time, and compute a compact data structure, that, given any set F of at most k vertices (or edges), efficiently computes vertices reachable from source in the graph GF. We show that for any set F of size k, our algorithm takes O(2^k n) time. We also consider reachability questions over a set of vertex-pairs in VxV. As an application, we obtain a fault tolerant algorithm for computing SSCs (strongly connected components) of a graph after k failures.
Bio:Keerti Choudhary is a postdoctoral fellow at the Weizmann Institute of Sciences, Israel, hosted by Prof. David Peleg. Prior to this, she completed her PhD in Computer Science from IIT Kanpur in August 2017, where she was advised by Prof. Surender Baswana. She is broadly interested in problems in graph theory and algorithms.
- Prophet Problems and Posted Prices
Abstract:Speaker: Ashish Chiplunkar
Date: 2019-02-04 Time: 12:00:00 (IST) Venue: SIT Building #001 The prophet problem is one of the central problems in optimal stopping theory. This problem and its related problems model the task of selling items to an online sequence of buyers. In the prophet problem, we are given the distributions of a sequence of independent random variables. We are then presented the realizations of the random variables in an online manner, and we must select one of them as soon as it is presented. When the sequence of random variables is adversarial, a celebrated result called the prophet inequality states that an algorithm can get, in expectation, at least half the expected maximum of the random variables. In the prophet secretary problem, the arrival sequence is uniformly random (but the distributions are adversarial). I will present an algorithm for prophet secretary which, in expectation, obtains at least (1-1/e+1/400) times the expected maximum. This improves the previously known (1-1/e) factor, which was believed to be optimal for natural reasons.
Bio:.
- Universal Access to All Information
Abstract:Speaker: Carl Malamud
Date: 2019-01-31 Time: 12:00:00 (IST) Venue: Bharti Building #501 For the past 30 years, Carl Malamud has helped place more than 5 crore pages of new information on the Internet. He'll discuss how he helped place large government databases, such as the U.S. Patent Database, on the Internet for free access over the objections of government. He'll also discuss his work in India over the past few years, running a collection of 4 lakh books called the Public Library of India and his efforts to make all 19,000 Indian Standards available. He will also discuss broader challenges, such as building a true public library of India and the moral imperative to make available the full scholarly corpus of journal articles for text and data mining. Carl will discuss universal access to information as the great promise of our times and the challenges we face in making that promise a reality.
Bio:Carl Malamud is founder of Public Resource, a California-based nonprofit organization that does extensive work on the Internet. He is the author of 9 professional reference books about computer networks and databases. In 1993, he ran the first radio station on the Internet. Carl has been a Visiting Professor at the MIT Media Lab and is the recipient of the Pioneer Award from the Electronic Frontier Foundation (EFF) and the Berkman Award from Harvard University "for his outstanding contributions to the Internet's impact on society."
Background Reading:
1. Interview: 'This Little USB Holds 19,000 Indian Standards. Why Should it Not Be Made Public?' https://thewire.in/tech/interview-little-usb-holds-19000-indian-standards-not-made-public
2. Op-Ed: Who May Swim in the Ocean of Knowledge? https://thewire.in/education/who-may-swim-in-the-ocean-of-knowledge - Universal Access to All Information
Abstract:Speaker: Dr Carl Malamud
Date: 2019-01-31 Time: 12:00:00 (IST) Venue: Bharti Building #501 For the past 30 years, Carl Malamud has helped place more than 5 crore pages of new information on the Internet. He'll discuss how he helped place large government databases, such as the U.S. Patent Database, on the Internet for free access over the objections of government. He'll also discuss his work in India over the past few years, running a collection of 4 lakh books called the Public Library of India and his efforts to make all 19,000 Indian Standards available. He will also discuss broader challenges, such as building a true public library of India and the moral imperative to make available the full scholarly corpus of journal articles for text and data mining. Carl will discuss universal access to information as the great promise of our times and the challenges we face in making that promise a reality.
Bio:Carl Malamud is founder of Public Resource, a California-based nonprofit organization that does extensive work on the Internet. He is the author of 9 professional reference books about computer networks and databases. In 1993, he ran the first radio station on the Internet. Carl has been a Visiting Professor at the MIT Media Lab and is the recipient of the Pioneer Award from the Electronic Frontier Foundation (EFF) and the Berkman Award from Harvard University "for his outstanding contributions to the Internet's impact on society."
Background Reading:
1. Interview: 'This Little USB Holds 19,000 Indian Standards. Why Should it Not Be Made Public?' https://thewire.in/tech/interview-little-usb-holds-19000-indian-standards-not-made-public
2. Op-Ed: Who May Swim in the Ocean of Knowledge? https://thewire.in/education/who-may-swim-in-the-ocean-of-knowledge - Extreme-Efficiency Computing
Abstract:Speaker: Raghav Pothukuchi
Date: 2019-01-23 Time: 12:00:00 (IST) Venue: Bharti Building #501 Computing is becoming ubiquitous and personal, powered by a variety of systems ranging from wearables to datacenters. However, this growth also leads us into an era of resource-constrained computing, where resources such as energy or storage often seem insufficient, and many constraints like temperature and power are imposed on the computation. My vision is to build Extreme Efficiency Computing (EEC) systems that are essential to deliver the desired Quality of Service (QoS), throughput and responsiveness. EEC systems avoid the undesirable consequences of inefficiency in this environment - loss of information, money and even human life. Currently, efficiency is sought through specialized processing, reconfigurable fabrics, and typically ad hoc control policies in the hardware, Operating System (OS) and application. However, results with ad hoc designs are often disappointing.
In my talk, I will describe my research contributions towards EEC so far and future directions. This is interdisciplinary work spanning Computer System Architecture, Control Theory, Machine Learning and Distributed Systems (Cloud and High-Performance Computing) design. I use theoretically solid techniques from Control Theory and Machine Learning to address the pervasive inefficiency in multiple layers of the computing stack. Using experiences from prototypes in research and industrial settings, I will present the challenges in this field and the effectiveness of the systematic control methods I have been developing.
Bio:Raghavendra (Raghav) is a PhD candidate at the University of Illinois, where he is advised by Prof. Josep Torrellas. His research is on building Extreme-Efficiency Computing systems. His work spans the areas of Computer Architecture, Control Theory, Machine Learning, OS, and Distributed System Design. He is a winner of the W. J. Poppelbaum Award by the Dept. of CS at Illinois for architecture design creativity, a Mavis Future Faculty Fellowship at Illinois, and an ACM SRC Competition at PACT'17. He interned at AMD and his modular control design is being considered for upcoming heterogeneous systems. He received his Masters in CS from Illinois in 2014 and worked at Nvidia before graduate school. He has a Bachelors (Hons.) in Electrical and Electronics Engineering from the Birla Institute of Technology & Science (BITS) Pilani, India where he received the university gold medal on graduation.
http://pothukuchi.web.illinois.edu/ - ABATe: Automatic Behavioral Abstraction Technique to Detect Anomalies in Smart Cyber-Physical Systems
Abstract:Speaker: Sandeep Nair Narayanan
Date: 2019-01-22 Time: 14:00:00 (IST) Venue: CSE, Seminar Room Detecting anomalies and attacks in smart cyber-physical systems are of paramount importance owing to their growing prominence in controlling critical systems. However, this is a challenging task due to the heterogeneity and variety of components of a CPS, and the complex relationships between sensed values and potential attacks or anomalies. Such complex relationships are results of physical constraints and domain norms which exists in many CPS domains. In this paper, we propose ABATe, an Automatic Behavioral Abstraction Technique based on Neural Networks for detecting anomalies in smart cyber-physical systems. Unlike traditional techniques which abstract the statistical properties of different sensor values, ABATe learns complex relationships between event vectors from normal operational data available in abundance with smart CPS's and uses this abstracted model to detect anomalies. ABATe detected more than 88% of attacks in the publicly available SWaT dataset featuring data from a scaled down sewage water treatment plant with a very low false positive rate of 1.3%. We also evaluated our technique's ability to capture domain semantics and multi-domain adaptability using a real-world automotive dataset, as well as a synthetic dataset.
Bio:Sandeep Nair Narayanan, Ebiquity Lab University of Maryland Baltimore County.
- Extracting and Labeling Information Facets
Abstract:Speaker: Mouna Kacimi
Date: 2019-01-22 Time: 12:00:00 (IST) Venue: SIT Building #001 Extracting information can be complex in advanced domains such as news, where events typically come with different facets including reasons, consequences, purposes, involved parties, and related events. The main challenge consists in first finding the set of facets related to each fact, and second tagging those facets to the relevant category. In this talk, I will present StuffIE, a fine-grained information extraction approach which is facet-centric. The proposed approach (1) breaks down triples having complex arguments into a set of facts and facets (2) provides nested relations, and (3) performs semantic tagging of the extracted information using machine learning strategies.I will also present in the talk the motivations and the applications of our work to give an overall picture of my research interests.
Bio:Mouna Kacimi is an assistant professor at UniBZ. She received her doctoral degree in computer science in 2007 from the University of Bourgogne, France. Afterwards, she had spent 3 years as a post-doc at the Databases and Information Systems group led by Gerhard Weikum at the Max Planck Institute for Informatics. Her research interests focus on enhancing search capabilities of information retrieval systems by developing new information extraction and text mining strategies.
- Improving Conversational Interactions at Alexa
Abstract:Speaker: Rahul Goel, Amazon
Date: 2019-01-21 Time: 15:30:00 (IST) Venue: Bharti Building #501 Digital Assistants are becoming commonplace. In this talk, we go over how do these systems work and what are some of the challenges which are being solved to make them better. One of the components in such a system is the spoken language understanding (SLU) component. Typically SLU systems factor the language understanding problem into intent classification and slot tagging. This has an inherent limitation for more complex linguistic phenomenon like coordination where the user expresses multiple intents in a statement.
In the latter half of the talk, we present 2 ways in which we can add structure to SLU. First, we talk about incorporating an external parser to solve coordination if we want to extend a legacy system. Second, we pose SLU as a shallow semantic parsing problem which is also able to handle tasks like Question Answering. We also talk about solving the data sparsity issue by doing transfer learning between domains and by using techniques like delexicalization and copy mechanism.
Bio:Rahul Goel is a machine learning scientist at Alexa AI where he works on improving spoken language understanding and dialog systems. Many of his contributions are currently deployed in Alexa. His research interests include dialog systems, language understanding, deep learning, and social computing. Before joining Amazon he was a graduate student at Georgia Tech working with Dr. Jacob Eisenstein on computational social science. Prior to that, he completed his bachelors from IIT, Delhi.
- A Framework for Specifying and Reasoning about Computations
Abstract:Speaker: Prof. Gopalan Nadathur
Date: 2019-01-16 Time: 12:00:00 (IST) Venue: Bharti Building #501 The computational structure of formal systems is often described through syntax directed rules in the style of structural operational semantics. Such presentations are typically intended to help in prototyping and in proving properties of these systems. This talk will survey techniques that we have developed to provide rigorous, computer-based support for such activities. Specifically,
I will discuss the higher-order abstract syntax approach to treating binding structure concisely and declaratively, the use of enriched recursive relational specifications to encode rule based descriptions, a logic of definitions that includes (co-)induction principles for reasoning about such relational specifications and a generic quantifier called nabla that allows for the treatment of names in specifications and free variables that arise during computation. I will also outline a two level logic approach to reasoning about specifications. These ideas have led to several actual computer systems such as Teyjus, Bedwyr, and Abella that will be touched upon during the talk.
Bio:Prof. Computer Science and Engineering, University of Minnesota
- "Complexity theory, cryptography and tests for quantumness"
Abstract:Speaker: Prof. Umesh Vazirani, UC Berkeley
Date: 2019-01-15 Time: 16:00:00 (IST) Venue: Bharti Building #501 TBA
Bio:Prof. Umesh Vazirani is a professor of computer science at UC Berkeley. He is a leading authority on complexity, algorithms and on quantum computing.
- Explainable AI: Making Visual Question Answering Systems more Transparent
Abstract:Speaker: Prof. Raymond J. Mooney, University of Texas at Austin
Date: 2019-01-11 Time: 12:00:00 (IST) Venue: Bharti Building #501 Artificial Intelligence systems' ability to explain their conclusions is crucial to their utility and trustworthiness. Deep neural networks have enabled significant progress on many challenging problems such as visual question answering (VQA), the task of answering natural language questions about images. However, most of them are opaque black boxes with limited explanatory capability. The goal of Explainable AI is to increase the transparency of complex AI systems such as deep networks. We have developed a novel approach to XAI and used it to build a high-performing VQA system that can elucidate its answers with integrated textual and visual explanations that faithfully reflect important aspects of its underlying reasoning while capturing the style of comprehensible human explanations. Crowd-sourced human evaluation of these explanations demonstrate the advantages of our approach.
Bio:Raymond J. Mooney is a Professor in the Department of Computer Science at the University of Texas at Austin. He received his Ph.D. in 1988 from the University of Illinois at Urbana/Champaign. He is an author of over 160 published research papers, primarily in the areas of machine learning and natural language processing. He was the President of the International Machine Learning Society from 2008-2011, program co-chair for AAAI 2006, general chair for HLT-EMNLP 2005, and co-chair for ICML 1990. He is a Fellow of the American Association for Artificial Intelligence, the Association for Computing Machinery, and the Association for Computational Linguistics and the recipient of best paper awards from AAAI-96, KDD-04, ICML-05 and ACL-07.
- Collusion Resistant Traitor Tracing from Learning with Errors
Abstract:Speaker: Rishab Goyal (UT Austin)
Date: 2019-01-10 Time: 15:00:00 (IST) Venue: SIT Building #001 A long-standing open problem in cryptography has been to build traitor tracing systems with compact ciphertexts. In this work we provide a traitor tracing scheme in which the ciphertext size grows polynomially in log(N) where N is the number of users and prove it secure under the Learning with Errors (LWE) assumption. This is the first traitor tracing scheme with such parameters provably secure from a standard assumption.
In a traitor tracing system, an authority runs a setup algorithm that takes as input the number, N, of users in the system. The setup outputs a public key PK, master secret key MSK, and N user keys (SK_1, ... , SK_N). The system has an encryption algorithm that uses the public key PK to create a ciphertext for a message M that is decryptable by any of the N user keys, but where the message will be hidden from any user who does not have access to the keys. Finally, suppose that some subset S subseteq [N] of users collude to create a pirate decoding box D which is capable of decrypting ciphertexts with some non-negligible probability. The tracing property of the system states that there exists a special algorithm Trace, which given the master secret key MSK and only oracle access to D, outputs a set of users T where T contains at least one user from the colluding set S and no users outside of S. The problem of traitor tracing was introduced by Chor, Fiat, and Naor in 1994.
In the talk, I will discuss where the hardness of traitor tracing stems from, and describe our results which is basically conceiving a novel approach to building traitor tracing that starts with a new form of Functional Encryption that we call Mixed FE.
I won't assume any prior knowledge of traitor tracing or related cryptographic primitives. This is joint work with Venkata Koppula and Brent Waters.
Full paper: https://eprint.iacr.org/2018/346
Bio:Rishab is a CSE, IIT Delhi graduate and is currently pursuing PhD at University of Texas at Austin (http://www.cs.utexas.edu/~rgoyal/)
- Symbolic Polynomial Evaluation.
Abstract:Speaker: Prof. Deepak Kapur
Date: 2019-01-09 Time: 15:30:00 (IST) Venue: Bharti Building #501 -
Bio:Deepak Kapur is a Honorary Professor in our Department, and is a professor and former chair of the CS Department at University of New Mexico, Albuquerque. He is visiting us till this weekend.
- Memory Defenses - the Elevation from Obscurity to Headlines
Abstract:Speaker: Prof. Rajeev Balasubramaniam, University of Utah
Date: 2019-01-08 Time: 15:00:00 (IST) Venue: SIT Building #001 ecent attacks like Meltdown and Spectre have highlighted that modern processors are likely being shipped with latent vulnerabilities that are impossible to anticipate. Some of these vulnerabilities can be addressed with various hardware defenses. Meltdown and Spectre may have finally pushed these defenses from the shadows of academia into possible commercial reality.
This talk will describe three primary vulnerabilities in the memory system, and efficient hardware defenses to address each of these vulnerabilities. The first vulnerability is leakage of a program's memory intensity through memory controller timing channels. The second is leakage of a program's memory access pattern through exposed DDR buses. The third is a violation of memory integrity by a malicious cloud operator or malicious OS. In fact, modern Intel SGX systems already have support for guaranteed memory integrity and we show how the performance of that solution can be improved
by nearly 4X.
Bio:Rajeev Balasubramonian is a Professor at the School of Computing, University of Utah. He received his B.Tech in Computer Science and Engineering from the Indian Institute of Technology, Bombay in 1998. He received his MS (2000) and Ph.D. (2003) degrees from the University of Rochester. His primary research interests include memory systems, security, and application-specific architectures, and his work appears regularly at the top architecture conferences. Prof. Balasubramonian is a recipient of a US National Science Foundation CAREER award, an IBM Faculty Partnership award, an HP Innovation Research Program award, an Intel Outstanding Research Award, various teaching awards at the University of Utah, and multiple best paper awards.
- Memory Defenses -- the Elevation from Obscurity to Headlines
Abstract:Speaker: Rajeev Balasubramaniam
Professor, School of Computing
University of UtahDate: 2019-01-08 Time: 15:00:00 (IST) Venue: SIT Seminar Room Recent attacks like Meltdown and Spectre have highlighted that modern
processors are likely being shipped with latent vulnerabilities that are
impossible to anticipate. Some of these vulnerabilities can be addressed
with various hardware defenses. Meltdown and Spectre may have finally
pushed these defenses from the shadows of academia into possible
commercial reality.
This talk will describe three primary vulnerabilities in the memory system,
and efficient hardware defenses to address each of these vulnerabilities.
The first vulnerability is leakage of a program's memory intensity through
memory controller timing channels. The second is leakage of a program's
memory access pattern through exposed DDR buses. The third is a violation
of memory integrity by a malicious cloud operator or malicious OS. In fact,
modern Intel SGX systems already have support for guaranteed memory
integrity and we show how the performance of that solution can be improved
by nearly 4X.
Bio:Rajeev Balasubramonian is a Professor at the School of Computing, University
of Utah. He received his B.Tech in Computer Science and Engineering from the
Indian Institute of Technology, Bombay in 1998. He received his MS (2000) and
Ph.D. (2003) degrees from the University of Rochester. His primary research
interests include memory systems, security, and application-specific
architectures, and his work appears regularly at the top architecture
conferences. Prof. Balasubramonian is a recipient of a US National Science
Foundation CAREER award, an IBM Faculty Partnership award, an HP Innovation
Research Program award, an Intel Outstanding Research Award, various teaching
awards at the University of Utah, and multiple best paper awards.
- Coding Space Theory and Optimization of 3D Cameras
Abstract:Speaker: Prof. Mohit Gupta, University of Wisconsin-Madison
Date: 2019-01-04 Time: 12:00:00 (IST) Venue: SIT Building #001 Time-of-flight (ToF) cameras have fast emerged as the preferred 3D imaging technique in several scientific and consumer applications, including robot navigation, motion capture, and human computer interfaces. Although the spatial resolution of ToF cameras continues to rise with advances in sensor technology, the depth resolution is fundamentally limited by the camera's coding functions. Imagine a user wearing an augmented reality headset equipped with a low-power ToF depth camera. For her to achieve a realistic immersive experience, the camera must measure the 3D structure of the surroundings with extreme resolution.
I will talk about our recent work on developing a formal coding theory of ToF cameras, and geometric optimization techniques for designing a new class of cameras with an order of magnitude higher depth resolution as compared to current cameras. These cameras can measure detailed 3D geometry (<100 microns) at long distances. Time permitting, I will also talk about cameras that can measure micro motions (e.g., detecting human pulse at long distances) using unconventional, but low-cost optics. These cameras can measure extremely subtle motions (< 10 microns), and, for the first time, track micro-motion for non-line-of-sight objects.
Bio:Mohit Gupta is an Assistant Professor of Computer Sciences at the University of Wisconsin-Madison. His research interests are in computer vision and computational imaging. He received B. Tech in Computer Science from IIT-Delhi, Ph.D. from the Robotics Institute, Carnegie Mellon University, and was a postdoctoral research scientist at Columbia University.
- Proof-of-Work without all the Work
Abstract:Speaker: Diksha Gupta
Date: 2019-01-04 Time: 12:00:00 (IST) Venue: SIT Building #001 The last decade has seen explosive growth in systems that are permissionless in that participants are free to join and leave at will. All such systems are open to the well- known Sybil attack, in which an adversary uses a large number of forged IDs to take control of the system. One of the most popular tools for Sybil defense is proof-of-work (PoW): IDs must periodically solve computational puzzles, thereby limiting the number of forged IDs.
Unfortunately, current PoW-based Sybil defenses suffer from a key weakness: the computational effort expended by the good IDs in solving puzzles must at least be equal to the computational effort of an attacker. We present the first algorithm to address this problem. Our algorithm is a Sybil defense that is asymmetric in the following sense: the good IDs spend at a rate that is asymptotically less than the attacker. In particular, if T is the rate of the attacker's spending, and J is the rate of joining good participants, then our algorithm spends at a rate O(J + √(JT) ).
Bio:Diksha is currently pursuing PhD at University of New Mexico under the supervision of Prof. Jared Saia. Her research is in the area of Distributed Computing, with primary focus on developing efficient algorithms to counter attacks on Distributed Systems. She is currently working on developing Proof-of-Work schemes where the amount of work required grows slowly as a function of the amount of work done by an attacker.
- Regular abstractions with applications to Infinite state verification
Abstract:Speaker: Dr Prakash Saivasan and TU Braunschweig
Date: 2019-01-03 Time: 12:00:00 (IST) Venue: SIT Building #001 Recursive programs (programs with the ability to make procedure calls) even over finite data domains are infinite state due to the unboundedness of the call stack. While sequential recursive programs can be modelled and verified via pushdown systems, verification in the presence of concurrency (programs with the ability to communicate), which is undecidable in general, remains an interesting and challenging problem. The focus of my research so far has been to address this problem via different techniques: under-approximations, accelerations and via regular abstractions. In this talk I will present one of our result on regular abstractions.
A regular abstraction is the approximation of an infinite state system as a finite automaton. For instance, one may approximate the behaviours (as a language) of a recursive program/pushdown system by its downward closure (i.e. the collection of all subwords of words in the language) and this is always a regular language. One may also disregard the order of letters in the words and consider the Parikh-image of the language. Again for recursive programs/ pushdown systems, this is representable by a finite state automaton.
I will explain the main ideas behind our results on computing regular abstractions for automata equipped with a counter. While such representations for pushdown systems involves an exponential blowup, we will see that the situation is significantly better for counter systems. It is polynomial for both upward and downward closures and quasi-polynomial for Parikh-image abstraction.
Time permitting, I will then show how to use the above result to carry out verification of quantitative properties for procedural programs. Our Quantitative logic provides the ability to express arithmetic constraints over the execution times of procedure invocations. In this logic one may express properties such as “within the execution of each invocation of a procedure P, the time spent in executing invocations of procedures Q and R is less than 15%”. I will also explain a second application of our result: in deciding the control state reachability problem for an under-approximation (bounded-stage runs) of concurrent recursive programs communicating via shared memory.