BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Computer Science and Engineering - ECPv6.13.0//NONSGML v1.0//EN CALSCALE:GREGORIAN METHOD:PUBLISH X-WR-CALNAME:Computer Science and Engineering X-ORIGINAL-URL:https://homecse.iitd.ac.in X-WR-CALDESC:Events for Computer Science and Engineering REFRESH-INTERVAL;VALUE=DURATION:PT1H X-Robots-Tag:noindex X-PUBLISHED-TTL:PT1H BEGIN:VTIMEZONE TZID:Asia/Kolkata BEGIN:STANDARD TZOFFSETFROM:+0530 TZOFFSETTO:+0530 TZNAME:IST DTSTART:20250101T000000 END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260520T160000 DTEND;TZID=Asia/Kolkata:20260520T170000 DTSTAMP:20261010T113105 CREATED:20260513T060719Z LAST-MODIFIED:20260513T060719Z UID:2520-1779292800-1779296400@homecse.iitd.ac.in SUMMARY:Algorithmic Behaviours in In-Context Learning by Dr. Aditya Gangrade DESCRIPTION:Venue: Bharti-501/ MS Teams \nAbstract: In-Context Learning (ICL) is a remarkable phenomenon whereby transformer-based LLMs can use data contained within their prompts to adapt their responses\, without changing their weights. This suggests that such models encode learning mechanisms. The recent literature has used statistical learning problems as a test-bed to investigate ICL\, and established that ICL can be realised for a wide range of function classes. However\, the mechanisms these models use to learn are poorly characterised. \nI will describe work on extracting and analysing learning algorithms embedded in the weights of transformers trained to perform ICL in two settings: linear-activation transformers for linear regression\, and softmax-activation transformers for linear classification. Through the former\, I will illustrate a high-level ‘simplify and validate’ strategy that allows extraction\, and through the latter\, I will describe a symmetry-driven strategy for evoking structure in these weights. In these settings\, we recover concrete iterative procedures that use existing ideas (Newton-Schulz; mean-shift methods) in new ways that are distinct from gradient descent. Further\, we show that transformers trained on variations of these problems implement modified versions of the same dynamics. This suggests that such models recover certain `stable’ algorithmic motifs\, and adapt them in response to problem structure. \nBased on work done jointly with Patrick Lutz\, Themistoklis Haris\, Arjun Chandra\, Hadi Daneshmand\, and Venkatesh Saligrama. \nBio:  Aditya Gangrade is a research scientist at the ECE department at Boston University. He obtained his Ph. D. in Systems Engineering from Boston University\, and previously held postdoctoral positions at Carnegie Mellon University and the University of Michigan. His research interests span theoretical and methodological aspects of machine learning\, with recent focus on safety in sequential decision making\, and on in-context learning phenomena. URL:https://homecse.iitd.ac.in/event/algorithmic-behaviours-in-in-context-learning-by-dr-aditya-gangrade/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260506T120000 DTEND;TZID=Asia/Kolkata:20260506T130000 DTSTAMP:20261010T113105 CREATED:20260503T085759Z LAST-MODIFIED:20260503T085759Z UID:2517-1778068800-1778072400@homecse.iitd.ac.in SUMMARY:Learning Hierarchical Control via Feasible Subgoal Prediction by Utsav Singh DESCRIPTION:Venue: SIT001 \nAbstract: Solving long-horizon tasks remains a central challenge in robotics because agents must explore efficiently\, assign credit over long time scales\, and act under sparse supervision. To address this\, hierarchical reinforcement learning (HRL) offers a promising alternative to flat reinforcement learning (RL) by enabling a high-level policy to propose subgoals and a low-level policy to execute them. However\, in practice\, HRL suffers from a fundamental issue: the higher level can propose subgoals that are infeasible for the lower level to achieve\, leading to training instability and sub-optimal performance. In this talk\, I will present my research around a central idea: hierarchy is effective only when its high-level decisions are grounded in the capabilities of the lower-level policies. I will discuss methods for training high-level policies to predict feasible subgoals by leveraging expert demonstrations\, preference-based feedback\, and visually grounded reward synthesis. Across challenging navigation and manipulation tasks\, these approaches improve training stability\, mitigate non-stationarity\, and enable agents to solve complex sparse-reward tasks in both simulation and real-world robotic settings. More broadly\, this work argues that building intelligent robotic agents that can solve long-horizon tasks is not just a matter of better planning\, but of ensuring that high-level decisions remain aligned with what lower-level policies can actually achieve. \n  \nBio: Utsav Singh received his Ph.D. from the Department of Computer Science and Engineering at the Indian Institute of Technology (IIT) Kanpur\, advised by Dr. Vinay P. Namboodiri and Dr. Sunil E. Simon. He will be joining the University of Central Florida as a Postdoctoral Researcher in June 2026\, working with Dr. Mubarak Shah and Dr. Amrit Singh Bedi. Prior to his Ph.D.\, he received his M.Tech. from IIT Kanpur. His research interests span hierarchical reinforcement learning (HRL)\, embodied intelligence\, and bilevel optimization\, with a current focus on enabling language-guided robotic control and improving reasoning in large language models (LLMs) via bilevel actor-critic frameworks. His research has been published at leading venues including ICML\, NeurIPS\, ICLR\, and AAAI. URL:https://homecse.iitd.ac.in/event/learning-hierarchical-control-via-feasible-subgoal-prediction-by-utsav-singh/ LOCATION:SIT 001\, Amar Nath and Shashi Khosla School of Information Technology\, IIT Delhi\, Hauz Khas\, New Delhi 110016\, India\, Delhi\, Delhi\, 110016\, India END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260430T163000 DTEND;TZID=Asia/Kolkata:20260430T173000 DTSTAMP:20261010T113105 CREATED:20260421T084518Z LAST-MODIFIED:20260421T084518Z UID:2507-1777566600-1777570200@homecse.iitd.ac.in SUMMARY:Physical reasoning in Minds\, Brains\, and Machines by Dr. Pramod RT DESCRIPTION:Online joining: https://teams.microsoft.com/meet/44299089959938?p=5wwN132pf54i4sVnyN \nAbstract: Successful engagement with the physical world involves perceiving the underlying structure\, predicting how things unfold\, and planning actions accordingly. This rich understanding and reasoning about our physical environment\, or ‘intuitive physics’\, develops early in infancy and is a core component of human cognition. While it seems easy for us to understand and interact in unfamiliar situations\, current machine learning systems are still far from achieving human-like generalizable performance. In this talk\, I will present: i) a set of non-invasive neuroimaging (functional Magnetic Resonance Imaging or fMRI) studies characterizing the brain basis of physical reasoning in humans\, ii) the first single-neuron level evidence for physical reasoning in the human brain using intracranial recordings\, and iii) ongoing work benchmarking latest AI models on various physical reasoning tasks. Together\, the results and methodology will help in not only understanding the neural mechanisms of physical reasoning but also discovering ways to bridge the reasoning gap between humans and AI. \nBio: RT Pramod is a computational cognitive neuroscientist at Massachusetts Institute of Technology whose research explores the neural and computational basis of physical scene understanding and reasoning. His work combines behavioral experiments\, computational modeling and neuroimaging to understand how humans perceive\, predict and plan in the world. Pramod obtained his Ph.D. from the Indian Institute of Science (IISc) working on compositional representations underlying visual object perception. His interests span visual cognition\, world models\, and the intersection of biological and machine intelligence. URL:https://homecse.iitd.ac.in/event/physical-reasoning-in-minds-brains-and-machines-by-dr-pramod-rt/ END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260424T120000 DTEND;TZID=Asia/Kolkata:20260424T130000 DTSTAMP:20261010T113105 CREATED:20260415T080647Z LAST-MODIFIED:20260415T080647Z UID:2495-1777032000-1777035600@homecse.iitd.ac.in SUMMARY:Learning assessment-aware brain representations from multimodal neuroimaging data by Dr. Ishaan Batta DESCRIPTION:Venue: SIT001 \nOnline joining: https://teams.microsoft.com/meet/48853006918605?p=788lqF84K1ykLq1rtg \nAbstract: Standard supervised learning on neuroimaging data optimizes for diagnostic prediction while yielding feature-level importance scores that lack network-level\, assessment-specific interpretability required for biomarker discovery; while unsupervised methods reduce data dimensions leading to loss of assessment-specific information. This talk presents frameworks developed towards addressing these gaps via biologically interpretable methodologies for neuroimaging data analysis. First\, a multimodal active subspace analysis framework to compute multiple salient directions that define the gradient space of a prediction function learned on brain features\, followed by repeated analysis to extract consistent and robust assessment-oriented subspace centers: compact multimodal representations of co-varying brain regions and functional connections maximally associated with a target clinical assessment. Second\, an interpretable deep learning framework\, constrained source-based salience\, that embeds active subspace learning and spatially constrained ICA directly into the saliency space of trained deep learning architectures\, producing network-level full-brain visualizations anchored around spatial brain templates. Lastly\, it will include some of the ongoing work on a conditional graph variational autoencoder that encodes static functional network connectivity in the brain into a structured latent space conditioned on demographic and cognitive variables\, enabling condition-specific reconstruction and identification of discriminative patterns of biological sex and fluid intelligence. Collectively\, these frameworks establish a principled methodology for learning brain representations that are simultaneously predictive\, network-interpretable\, and account for clinical observations. \nBio: Ishaan Batta is a postdoctoral research associate at the tri-institutional Center for Translational Research in Neuroimaging and Data Science (TReNDS)\, GSU/GAtech/Emory\, Atlanta\, USA. His research lies at the intersection of machine learning and neuroscience\, with a focus on developing interpretable representation learning methods for high-dimensional multimodal neuroimaging data to uncover biologically meaningful signatures of brain disorders and cognitive function.\nIshaan completed his Ph.D. in Electrical and Computer Engineering at the Georgia Institute of Technology (Georgia Tech)\, USA in 2023\, advised by Dr. Vince Calhoun. His doctoral and postdoctoral work has introduced a suite of novel frameworks spanning active subspace learning\, deep learning-based interpretation\, and conditional generative modeling\, aimed at ensuring both predictive performance as well as neurobiological interpretability in brain imaging studies. Prior to his graduate studies\, Ishaan received a dual degree (B.Tech. and M.Tech.) in Computer Science and Engineering from the Indian Institute of Technology Delhi (IIT Delhi) in 2017 URL:https://homecse.iitd.ac.in/event/learning-assessment-aware-brain-representations-from-multimodal-neuroimaging-data-by-dr-ishaan-batta/ LOCATION:SIT 001\, Amar Nath and Shashi Khosla School of Information Technology\, IIT Delhi\, Hauz Khas\, New Delhi 110016\, India\, Delhi\, Delhi\, 110016\, India END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260420T153000 DTEND;TZID=Asia/Kolkata:20260420T163000 DTSTAMP:20261010T113105 CREATED:20260418T084932Z LAST-MODIFIED:20260418T084932Z UID:2505-1776699000-1776702600@homecse.iitd.ac.in SUMMARY:Quantum Computing: Towards Advantage by Dhinakaran Vinayagamurthy DESCRIPTION:Venue: Bharti501 \nAbstract: This talk will provide a perspective on where we are at IBM Quantum in building useful quantum computers. There are two main pillars: developing a quantum computing platform that scales beyond classical computers\, and discovering algorithms that leverage the strengths of this platform to deliver state-of-the-art methods for solving hard problems. The talk will also provide an overview of how tools available in Qiskit can be leveraged for research and development. \nBio: Dhinakaran Vinayagamurthy is a Researcher in the newly formed Quantum Computing research group at the IBM Research India lab in Bangalore\, and the Engagement Manager for the IIT Madras-IBM Quantum partnership. My research interests are in quantum error mitigation\, cryptography and security. At IBM\, I have worked on projects around blockchain interoperability\, encrypted databases\, IBM Blockchain Transparent Supply and privacy-preserving machine learning. I am also an IBM Quantum Senior Ambassador. URL:https://homecse.iitd.ac.in/event/quantum-computing-towards-advantage-by-dhinakaran-vinayagamurthy/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260416T120000 DTEND;TZID=Asia/Kolkata:20260416T130000 DTSTAMP:20261010T113105 CREATED:20260415T080141Z LAST-MODIFIED:20260415T080141Z UID:2491-1776340800-1776344400@homecse.iitd.ac.in SUMMARY:The Landscape of Exact Round Complexity in Secure Multi-Party Computation by Prof. Arpita Patra DESCRIPTION:Venue: Bharti501\nAbstract: Secure Multi-Party Computation (MPC) is a central problem in cryptography\, often regarded as its holy grail. It enables a group of mutually distrusting data owners to jointly compute a function over their private inputs\, while revealing nothing beyond what is inherently implied by the output itself.\nRound complexity is one of the most fundamental efficiency measures in MPC\, capturing the minimal interaction required for secure computation. In this talk\, I will present a high-level overview of the evolution of research on round complexity in MPC and place my own contributions within this evolving landscape.\n\nBio: Arpita Patra is a Professor of Computer Science at the Indian Institute of Science (IISc). Her research focuses on cryptography\, with particular emphasis on the theoretical foundations and practical implementations of secure computation protocols. She has authored over 100 publications\, and her work has been recognized through numerous honors\, including the Prof. S. K. Chatterjee Award for Outstanding Woman Researcher/Industry Leader (IISc\, 2023)\, the Google Privacy Research Faculty Award (2023)\, the J.P. Morgan Chase Faculty Award (2022)\, the SONY Faculty Innovation Award (2021)\, the Google Research Award (2020)\, the NASI Young Scientist Platinum Jubilee Award (2018)\, the SERB Women Excellence Award (2016)\, and the INAE Young Engineer Award (2016). She is affiliated with leading scientific academies\, including the Indian Academy of Sciences (IAS)\, the Indian National Academy of Engineering (INAE)\, and The World Academy of Sciences (TWAS). She has coauthored two academic textbooks: Secure Multiparty Computation against Passive Adversaries (Springer\, 2023) and Fault Tolerant Distributed Consensus in Synchronous Networks (Springer\, 2025). URL:https://homecse.iitd.ac.in/event/the-landscape-of-exact-round-complexity-in-secure-multi-party-computation-by-prof-arpita-patra/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260416T100000 DTEND;TZID=Asia/Kolkata:20260416T110000 DTSTAMP:20261010T113105 CREATED:20260415T081035Z LAST-MODIFIED:20260415T081035Z UID:2497-1776333600-1776337200@homecse.iitd.ac.in SUMMARY:Challenges in scaling memory bandwidth in modern SoCs by Nithya Bashyam DESCRIPTION:Venue: Bharti501\nAbstract: Memory bandwidth scaling has emerged as a critical bottleneck as compute capabilities continue to grow faster than DRAM performance. This talk examines why higher memory frequencies do not directly translate to higher delivered bandwidth\, focusing on bandwidth efficiency rather than peak bandwidth. It discusses key constraints arising from DRAM organization\, timing parameters\, power\, reliability and security\, and highlights the roles of memory controllers\, SoC architecture\, and workload behavior. The talk provides a framework to understand the architectural trade‑offs involved in sustaining memory bandwidth scaling in modern SoCs.\n\n\nBiography: A Principal Engineer in Client SoC Performance Architecture at Intel\, Nithya Bashyam holds an M.Tech. in Electronics Design from CEDT\, IISc. He joined Intel in 2004\, initially working on front‑end design for two years\, before moving in 2006 to the CPU SoC Architecture team. He contributed to the industry’s first CPU with integrated graphics and memory controller on Sandy Bridge\, released in 2011\, followed by work on a few generations of server CPUs. Since Skylake (2015)\, he has focused on client CPUs\, working across the core‑to‑memory path including cache\, ring\, memory controllers\, and aspects of the I/O subsystem. He has led memory subsystem performance for several generations of client SoCs and led SoC performance for Lunar Lake (2024) and Panther Lake (2025). URL:https://homecse.iitd.ac.in/event/challenges-in-scaling-memory-bandwidth-in-modern-socs-by-nithya-bashyam/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260415T160000 DTEND;TZID=Asia/Kolkata:20260415T170000 DTSTAMP:20261010T113105 CREATED:20260415T080415Z LAST-MODIFIED:20260415T080415Z UID:2493-1776268800-1776272400@homecse.iitd.ac.in SUMMARY:State of Confidential Computing by Dr. Kapil Vaswani DESCRIPTION:Venue: Bharti501 \nAbstract: Over the last decade\, confidential computing has emerged as an advanced security and privacy primitive that can fundamentally change the nature of trust that we place in digital services. In this talk\, we will take a journey through the evolution of confidential computing and understand its current state and open problems that will need to be solved. Towards the end\, I will introduce SPARC\, a new research center working at the intersection of AI and security\, and our view of the opportunities that confidential computing creates for building the next generation of AI systems. \nBio: Kapil Vaswani is the founder of SPARC. He is a security researcher with 18 years of experience in research spanning computer architecture\, programming languages\, systems\, hardware security\, and AI security. At Microsoft Research\, he led research on problems in confidential computing\, privacy-preserving AI\, database security\, supply chain security\, and formally verified hardware. His research has been published in many top tier conferences such as POPL\, PLDI\, ASPLOS\, MICRO\, OSDI\, Oakland and USENIX Security. His work directly led to several products at Microsoft including SQL Server Always Encrypted\, Azure Confidential GPU VMs\, Azure AI Confidential Inferencing\, Azure Confidential Clean Rooms\, and Ad Selection API.  In his capacity as a volunteer with iSPIRT\, he has led the development of the DEPA Inferencing and Training Framework. ​Kapil Vaswani has a master’s and PhD from the Indian Institute of Science\, Bangalore. URL:https://homecse.iitd.ac.in/event/state-of-confidential-computing-by-dr-kapil-vaswani/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260413T120000 DTEND;TZID=Asia/Kolkata:20260413T130000 DTSTAMP:20261010T113105 CREATED:20260407T073939Z LAST-MODIFIED:20260407T073939Z UID:2483-1776081600-1776085200@homecse.iitd.ac.in SUMMARY:Advancing Safe Multimodal Intelligence by Dr. Pritam Sarkar DESCRIPTION:Venue: Bharti-501/ MS Teams \nAbstract: Building artificial intelligence with meaningful real-world impact requires models that can understand and interact in both the virtual and physical world. This talk outlines four key capabilities necessary to achieve this: foundational world knowledge\, alignment with human values and expectations\, reasoning ability\, and the capacity to self-improve. The talk begins with an overview of our past contributions toward these capabilities\, with a particular focus on building foundational world knowledge in multimodal models and improving their alignment with human values and expectations. The next part of the talk delves into a fundamental challenge in current alignment approaches: the alignment tax. While existing methods aim to make models safer and more reliable\, they often degrade general capabilities—reducing response diversity and making models overly cautious or less useful. To address this\, we introduce Refined Regularized Preference Optimization (RRPO)\, a fine-grained alignment method. By penalizing only specific error tokens rather than entire responses\, RRPO mitigates harmful behaviors while simultaneously improving performance across diverse vision tasks. Finally\, the talk concludes with an overview of ongoing work on enabling stronger visual reasoning and outlines a research vision for developing machines that can adapt and improve over time\, fostering long-term usefulness and human–AI trust. \n\n \n \nBio: Pritam Sarkar is a Distinguished Postdoctoral Fellow at the Vector Institute and a Postdoctoral Research Fellow at the University of British Columbia\, where he works on multimodal AI\, especially with video\, image\, audio\, and language. He completed his PhD in September 2025 at Queen’s University\, Canada and during this time\, he was an intern at Google\, USA. His research has been recognized at leading venues including NeurIPS\, ICLR\, and AAAI\, with multiple Oral and Spotlight presentations. He received the IEEE Research Excellence Award in 2023 for his work on self-supervised learning. He actively serves the research community as an Area Chair and a Reviewer for leading conferences and journals such as NeurIPS\, CVPR\, and PAMI\, and is a strong proponent of open-source research. He is interested in developing safe and generalizable multimodal intelligence through algorithms that learn effectively with minimal human supervision. Find more: https://pritamsarkar.com/  \nHead Shot: https://pritamsarkar.com/assets/my_images/pp_square.jpg  URL:https://homecse.iitd.ac.in/event/advancing-safe-multimodal-intelligence-by-dr-pritam-sarkar/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260406T110000 DTEND;TZID=Asia/Kolkata:20260406T120000 DTSTAMP:20261010T113105 CREATED:20260404T150052Z LAST-MODIFIED:20260404T150052Z UID:2480-1775473200-1775476800@homecse.iitd.ac.in SUMMARY:Machine Learning under Adversaries: How Structure in Data Helps by Ambar Pal DESCRIPTION:Venue: SIT001 \nAbstract: This talk overviews recent results in the theoretical foundations of adversarially robust machine learning. Modern ML classifiers can fail spectacularly when subject to specially crafted input-perturbations\, called adversarial examples. On the other hand\, humans are quite robust for several tasks involving vision. Motivated by this contrast\, in the first part of this talk we will take a deeper dive into the question of when exactly adversarial examples can be avoided. We will see that a key property of the data distribution — localization on small volume subsets of the input space — characterizes whether any robust classifier exists. In the second part of this talk\, we will empirically instantiate these results for a few localized data distributions\, and demonstrate that utilizing such structure in data leads to practical classifiers that enjoy better provable robustness guarantees in several regimes. This talk is based on work at NeurIPS ’23\, ’20 and TMLR ’23\, ’24. \n  \nBio: Ambar Pal is a scientist at Amazon Responsible AI. He received his PhD in Computer Science from the Johns Hopkins University. His current research is in the foundations of robust machine learning where he develops the theory and practice of tools that closely utilize structure in data for robust machine learning. His research has been awarded the CPAL rising star award and fellowships from JHU and Amazon. URL:https://homecse.iitd.ac.in/event/machine-learning-under-adversaries-how-structure-in-data-helps-by-ambar-pal/ LOCATION:SIT 001\, Amar Nath and Shashi Khosla School of Information Technology\, IIT Delhi\, Hauz Khas\, New Delhi 110016\, India\, Delhi\, Delhi\, 110016\, India END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260316T120000 DTEND;TZID=Asia/Kolkata:20260316T130000 DTSTAMP:20261010T113105 CREATED:20260312T134932Z LAST-MODIFIED:20260312T134932Z UID:2437-1773662400-1773666000@homecse.iitd.ac.in SUMMARY:Logical explorations for security theory by Prof. Vaishnavi Sundararajan DESCRIPTION:Venue: Bharti 501 \nAbstract: Logics and proof theories play a large role in the formal study and analysis of systems\, especially for formal verification. The exact shape of the syntax and proof rules involved depend heavily on the systems being modelled and verified. In this talk\, we will introduce a logical syntax for communicated messages and an associated proof system originally used in the verification of cryptographic protocols\, dating back to a very robust model from 1983\, which captures even the operational aspects of today’s internet. We will show how to extend thissystem to be able to better handle protocols that involve the communication of certificates\, and investigate some of the various logical and algorithmic questions that manifest during this endeavour. URL:https://homecse.iitd.ac.in/event/logical-explorations-for-security-theory-by-prof-vaishnavi-sundararajan/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260313T120000 DTEND;TZID=Asia/Kolkata:20260313T130000 DTSTAMP:20261010T113105 CREATED:20260309T161051Z LAST-MODIFIED:20260309T161051Z UID:2434-1773403200-1773406800@homecse.iitd.ac.in SUMMARY:Genteel-Negotiator: LLM-enhanced mixture-of-expert-based reinforcement learning approach for polite negotiation dialogue by Dr. Mauajama Firdaus DESCRIPTION:Venue: SIT001 \nAbstract : Developing intelligent negotiation dialogue systems that promote fair and sustainable outcomes is crucial for advancing automated negotiation for social good. Since effective negotiation requires balancing cooperation and competition while maintaining respect\, we propose GENTEEL-NEGOTIATOR\, a polite negotiation dialogue system for tourism and e-commerce domains. We introduce NEGOCHAT\, a tourism negotiation dataset\, and enrich it along with the Integrative Negotiation Dataset (IND) using diverse negotiation strategies. Built on an LLM-enhanced Mixture-of-Experts reinforcement learning framework\, the model incorporates dedicated experts for negotiation\, politeness\, and coherence\, guided by a reward function capturing strategy alignment\, politeness\, coherence\, and engagingness. Extensive automatic and human evaluations demonstrate its effectiveness in generating polite\, coherent\, and goal-oriented negotiation responses. \n\nBio: Dr. Mauajama Firdaus is an Assistant Professor in the Department of Computer Science and Engineering at the Indian Institute of Technology (ISM) Dhanbad. She completed her Ph.D. in Computer Science & Engineering from IIT Patna and pursued postdoctoral research at the University of Alberta\, Canada. Her research expertise lies in Natural Language Processing\, Multimodal and Multilingual AI\, Dialogue Systems\, Explainable AI\, and Affective Computing. Her work focuses on building empathetic\, polite\, and emotion-aware conversational AI systems\, with applications in social good\, mental health\, customer care\, and multilingual settings. She has published extensively in top-tier journals and conferences including IEEE\, ACM\, AAAI\, ACL\, EMNLP\, and Information Fusion\, and holds a US patent in spoken language understanding. She also serves as Associate Editor for reputed international journals published by Elsevier. URL:https://homecse.iitd.ac.in/event/genteel-negotiator-llm-enhanced-mixture-of-expert-based-reinforcement-learning-approach-for-polite-negotiation-dialogue-by-dr-mauajama-firdaus/ LOCATION:SIT 001\, Amar Nath and Shashi Khosla School of Information Technology\, IIT Delhi\, Hauz Khas\, New Delhi 110016\, India\, Delhi\, Delhi\, 110016\, India END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260310T090000 DTEND;TZID=Asia/Kolkata:20260310T100000 DTSTAMP:20261010T113105 CREATED:20260302T102353Z LAST-MODIFIED:20260302T102353Z UID:2429-1773133200-1773136800@homecse.iitd.ac.in SUMMARY:Integration of Structured Reasoning and Data-driven Learning for Acting and Planning by Dr. Sunandita Patra DESCRIPTION:Teams Link: MS Teams\nAbstract: The talk will focus on enabling autonomous actors\, such as digital agents or robots\, to take deliberative actions towards achieving their long horizon goals\, in the face of uncertainty and dynamic events. Today\, dynamic events or failures often require human intervention\, system restarts\, retraining\, or redesign\, for example\, when robots get stuck in dead ends or digital agents collapse under unanticipated events. Existing methods either rely on rule-based reasoning\, where it is extremely difficult for human experts to compile and maintain a complete set of rules\, or black-box machine learning models that need to be trained extensively for individual tasks. Neither approach is well suited when actors are operating in a dynamically changing environment. This work aims to overcome these limitations by creating integrated planning and learning algorithms that are practical to be executed in real-world environments (spanning finance\, robotics and cybersecurity) by incorporating within a single framework: (a) deliberative acting\, (b) online planning\, and (c) data-driven learning from the actor’s experiences.\n\n\nBy integrating structured reasoning with data-driven learning\, the goal of this research is to push towards the next generation of autonomous systems\, general-purpose agentic AI that can plan and act deliberately across a set of diverse tasks and domains.\n\n \nBio: Sunandita Patra is a Research Lead at J. P. Morgan AI Research\, Chicago\, USA. Her research interests include the integration of acting\, planning\, and machine learning\, focusing on finance\, cybersecurity\, and robotics domains. She completed her PhD and PostDoc in Computer Science at the University of Maryland\, College Park\, and holds BTech and MTech degrees in Computer Science and Engineering from IIT Kharagpur.  Her work received the Best Student Paper Honorable Mention Award at ICAPS 2020\, and the Best Student Paper Finalist Award (Top 3) at ECMR 2025. For more details\, please visit https://sunanditapatra.wixsite.com/camp URL:https://homecse.iitd.ac.in/event/integration-of-structured-reasoning-and-data-driven-learning-for-acting-and-planning-by-dr-sunandita-patra/ END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260306T120000 DTEND;TZID=Asia/Kolkata:20260306T130000 DTSTAMP:20261010T113105 CREATED:20260303T103646Z LAST-MODIFIED:20260303T103646Z UID:2432-1772798400-1772802000@homecse.iitd.ac.in SUMMARY:Neural Circuit Discovery via Representation and Dynamics by Savik Kinger DESCRIPTION:Abstract: Neuroscience and AI share a bottleneck: while one can build (artificial) or record (biological) complex networks\, we struggle to explain their functional circuitry; i.e.\, how they compute. In this talk I use whole-brain recordings from C. elegans\, a canonical neurobiological system\, as a concrete testbed for “circuit interpretability.” I then introduce two complementary inference approaches for turning high-dimensional activity data into mechanistic structure. Approach 1 treats circuit discovery as a representation problem: learn time-varying functional structure and uncover recurring\, stimulus-dependent modules rather than a single static connectivity map. Approach 2 treats circuit discovery as a dynamics problem: go beyond correlation to estimate directed\, time-lagged influence—i.e.\, which units appear to drive others and over what delays—using modern score-based generative modeling ideas. I will show how these ML methods produce testable hypotheses for biologists and\, potentially\, offer new avenues for understanding complex networks in AI. \nBio: Savik Kinger is a PhD candidate in Computer Science at Yale University\, advised by Steven Zucker. His research focuses on developing methods to analyze biological and artificial neural networks\, integrating ideas from machine learning\, dynamical systems\, and causal inference. He received Bachelor’s degrees in Math and Computer Science from Columbia University. He is supported by a Nathan Hale fellowship. URL:https://homecse.iitd.ac.in/event/neural-circuit-discovery-via-representation-and-dynamics-by-savik-kinger/ LOCATION:SIT 113\, Amar Nath and Shashi Khosla School of Information Technology\, Indian Institute of Technology\, Delhi\, Hauz Khas\, New Delhi\, Delhi\, 110016\, India CATEGORIES:Seminars END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260219T110000 DTEND;TZID=Asia/Kolkata:20260219T120000 DTSTAMP:20261010T113105 CREATED:20260217T085749Z LAST-MODIFIED:20260217T085749Z UID:2426-1771498800-1771502400@homecse.iitd.ac.in SUMMARY:TESSERA: Programming Petabytes of Earth Observations using Foundation Models by Prof. Anil Madhavapeddy DESCRIPTION:Venue. Bharti 501\n\nAbstract. We present TESSERA\, a pixel-wise foundation model for multi-modal (Sentinel-1/2) earth observation time series that learns robust\, label-efficient embeddings.  Our goal with TESSERA is to make manipulating global satellite intelligence as easy as LLMs did for natural language! Towards this we release global\, annual\, 10m\, pixel-wise embeddings together with open weights and code and lightweight adaptation heads\, providing practical tooling for large-scale retrieval and inference at planetary scale.  As with any good foundation model\, there are a staggering array of downstream tasks which can benefit. TESSERA embeddings deliver state-of-the-art accuracy with high label efficiency across diverse classification\, segmentation\, and regression tasks.\n\nIn this talk\, I’ll take you through an array of problems our users are applying it to\, ranging from the ecological to the urban to the temporal. By the end of the talk\, we aim to have you identify a seemingly impossible spatial problem that is now within range to solve yourself using our easy-to-install Python package\, geotessera. Bring your favourite coding agents!\n\nSpeaker: Anil Madhavapeddy\, Professor of Planetary Computing\, University of Cambridge URL:https://homecse.iitd.ac.in/event/tessera-programming-petabytes-of-earth-observations-using-foundation-models-by-prof-anil-madhavapeddy/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260212T120000 DTEND;TZID=Asia/Kolkata:20260212T130000 DTSTAMP:20261010T113105 CREATED:20260204T131611Z LAST-MODIFIED:20260204T131643Z UID:2416-1770897600-1770901200@homecse.iitd.ac.in SUMMARY:Abstractions for expressive\, extensible\, and scalable root cause analysis by Vipul Harsh DESCRIPTION:Venue: Bharti501 \nAbstract: Modern Internet-scale services must identify and mitigate customer-impacting incidents quickly. Despite the development of many Root Cause Analysis (RCA) algorithms—including recent LLM-assisted solutions—existing approaches struggle with the “long tail” of novel failure modes and the sheer scale of telemetry. In this talk\, I argue that the path forward requires a paradigm shift from developing point-solution algorithms to a systems-first approach. I will introduce MoCE: a Mixture-of-Experts (MoE) framework that provides high-level abstractions for failure diagnosis. This framework enables developers to express complex troubleshooting logic succinctly using MoCE’s domain specific operators while providing the underlying systems support for scalable telemetry processing. Finally\, I will briefly describe how these abstractions empower reliable\, autonomous agents to perform interactive diagnosis and discuss ongoing and promising future work based on these ideas. \nBio: Vipul Harsh is a postdoctoral researcher at Conviva with Vyas Sekar and Hui Zhang and a visiting researcher at Carnegie Mellon University (CMU). His research lies at the intersection of networked systems\, AI\, and theory. His works span design of reliable agents for interactive troubleshooting\, systems for failure diagnosis in networked systems\, datacenter topology\, distributed monitoring\, and parallel algorithms. His research has been published in top-tier CS conferences (SIGCOMM\, NSDI\, SPAA among others) and his projects have been adopted into real-world products. He completed his Ph.D. from UIUC where he worked with Brighten Godfrey and holds an undergraduate degree from IIT Bombay. His thesis was nominated by UIUC for the ACM SIGCOMM dissertation award. URL:https://homecse.iitd.ac.in/event/2416/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260206T160000 DTEND;TZID=Asia/Kolkata:20260206T170000 DTSTAMP:20261010T113105 CREATED:20260206T095221Z LAST-MODIFIED:20260206T095221Z UID:2419-1770393600-1770397200@homecse.iitd.ac.in SUMMARY:Molecular Machine Learning for Chemical Catalysis by Dr. Sukriti Singh DESCRIPTION:Venue: SIT001\nAbstract: The development of new reaction methodology could become a tedious task demanding both time and resources. The application of machine learning (ML) approaches for reaction optimization and prediction can make a significant impact on efficient exploration of the high-dimensional chemical space. But the direct adaptation of ML as used in well-developed domains\, is unlikely to succeed in reaction discovery. Some of the challenges stem from ineffective featurization of the molecular space\, unavailability of quality data and its distribution. Given these backgrounds\, rendering ML tools conducive for reactions is an exciting as well as challenging endeavor at the same time.\nIn this talk\, I will present molecular machine learning strategies specifically designed for small-data reaction discovery\, typically involving only hundreds to a few thousand data points. I will first discuss feature engineering approaches based on quantum-chemically derived physical organic descriptors\, illustrated through catalytic asymmetric hydrogenation of imines and alkenes for predicting enantioselectivity. I will then introduce feature learning methods that learn molecular representations directly from data and demonstrate their effectiveness in predicting reaction yield and enantioselectivity across diverse catalytic transformations.\n\nTo address data scarcity\, I will describe a transfer learning framework in which a chemical language model is trained on large number of molecules and fine-tuned on a focused library of desired reactions. Finally\, I will briefly introduce a meta-learning workflow that leverages literature-derived reaction data to identify shared reaction features\, enabling accurate outcome prediction with only a few experimental examples. Overall\, this talk will highlight how technically sound deployment of molecular machine learning tools can guide reaction development and help us get closer to sustainable practices by reducing the number of heuristic and empirical steps.\n\nBio: Sukriti Singh received her M.Sc. and Ph.D. degrees from the Department of Chemistry\, IIT Bombay in 2022 under the supervision of Prof. Raghavan B. Sunoj. She received the Naik and Rastogi Award for Excellence in Ph.D. Research. She was a postdoctoral research associate with Prof. J. M. Hernandez-Lobato at the Department of Engineering\, University of Cambridge. Her research interests involve density functional theory studies of catalytic reactions and developing machine learning methods to tackle low-data situations aimed at accelerating the exploration of chemical reaction space of high contemporary interest. URL:https://homecse.iitd.ac.in/event/molecular-machine-learning-for-chemical-catalysis-by-dr-sukriti-singh/ LOCATION:SIT 001\, Amar Nath and Shashi Khosla School of Information Technology\, IIT Delhi\, Hauz Khas\, New Delhi 110016\, India\, Delhi\, Delhi\, 110016\, India END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260205T120000 DTEND;TZID=Asia/Kolkata:20260205T130000 DTSTAMP:20261010T113105 CREATED:20260131T075508Z LAST-MODIFIED:20260131T075508Z UID:2410-1770292800-1770296400@homecse.iitd.ac.in SUMMARY:Approximately Packing Dijoins Via Nowhere-Zero Flows by Dr. Ravi DESCRIPTION:Venue: Bharti501 \nAbstract: In a digraph\, a dicut is a cut where all the arcs cross in one direction. A dijoin is a subset of arcs that intersects each dicut. Woodall conjectured in 1976 that in every digraph\, the minimum size of a dicut equals to the maximum number of disjoint dijoins. By building connections with nowhere-zero k-flows\, we prove that every digraph with minimum dicut size $\tau$ contains $\lfloor \tau/k \rfloor$ disjoint dijoins if the underlying undirected graph admits a nowhere-zero k-flow. \nJoint work with Gérard Cornuéjols (CMU) and Siyue Liu (CMU) \nSpeaker Bio: Dr. Ravi is the Vasantrao Dempo Professor of Operations Research and Computer Science at Carnegie Mellon University. His research is on models\, methods and applications of discrete optimization and their application to business and technological systems. He has published widely in diverse areas ranging from theoretical computer science to Operations and Marketing. In Computer Science\, his main research interests are in approximation algorithms and network optimization. URL:https://homecse.iitd.ac.in/event/approximately-packing-dijoins-via-nowhere-zero-flows-by-dr-ravi/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260204T120000 DTEND;TZID=Asia/Kolkata:20260204T130000 DTSTAMP:20261010T113105 CREATED:20260201T193547Z LAST-MODIFIED:20260201T193547Z UID:2413-1770206400-1770210000@homecse.iitd.ac.in SUMMARY:A new characterization of VNP via colored determinant by Dr. Prasad Chaugule DESCRIPTION:Venue: Bharti501 \nAbstract: Understanding the algebraic complexity class VNP through alternative characterizations is a central theme in algebraic complexity theory\, closely tied to the VP vs. VNP problem. While the permanent provides a canonical complete polynomial for VNP\, identifying natural and combinatorial variants that lead to new structural insights remains an important challenge.In this talk\, I will present a new characterization of VNP based on acombinatorial variant of the determinant\, which we call the colored determinant. This polynomial is defined as a signed sum over properly colored cycle covers of a directed graph\, where each cycle is required to be monochromatic. We show that the colored determinant is VNP-complete under p-projections over all fields\, thereby adding a new non-monotone VNP-complete polynomial family distinct from the permanent and previously studied determinant variants. \nUsing this polynomial\, we introduce a new computational model called the conditional stack branching program. Unlike standard stack branching programs\, this model allows the stack operation on an edge to depend on the current top of the stack. We show that this added conditional power is sufficient to increase expressiveness: a single-stack conditional stack branching program already characterizes VNP. This sharply contrasts with prior results\, where at least two stacks were required to capture VNP. \n  \nSpeaker Bio: Dr. Prasad Chaugule is a Post Doctoral fellow in the Theory Group at Department of Computer Science and Engineering\, IIT Delhi. His research lies in Arithmetic Circuit Complexity. He earned his Ph.D from IIT Bombay. URL:https://homecse.iitd.ac.in/event/a-new-characterization-of-vnp-via-colored-determinant-by-dr-prasad-chaugule/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260202T120000 DTEND;TZID=Asia/Kolkata:20260202T130000 DTSTAMP:20261010T113105 CREATED:20260115T072749Z LAST-MODIFIED:20260115T072749Z UID:2365-1770033600-1770037200@homecse.iitd.ac.in SUMMARY:Deep generative models for single-cell and spatial genomics by Ajita Shree DESCRIPTION:Speaker: Ms. Ajita Shree is a PhD student in the Department of Computer Science and Engineering at IIT Kanpur and will be joining EMBL-EBI\, UK\, as a postdoctoral researcher. \nAbstract: Recent advances in large-scale genomic assays\, including single-cell and spatial transcriptomics (ST)\, have provided unprecedented insights into the biological mechanisms underlying development\, disease\, and therapeutic response. However\, these datasets pose significant computational challenges. \nOne of the major challenges is the integration of heterogeneous single-cell datasets across donors\, time points\, and experimental conditions to create a unified resource for downstream analysis. In this talk\, I will present scDREAMER\, a novel deep generative model that can perform integration of multi-batch single-cell datasets in unsupervised\, semi-supervised and supervised settings. Using real benchmarking datasets\, we demonstrate that scDREAMER can overcome critical challenges including skewed cell type distribution among batches\, nested batch-effects\, large number of batches and enables integration of millions of cells across species. \nFurther\, I will discuss the emerging field of spatial genomics\, which enables the study of gene expression at a spatial level\, but its spot-level resolution poses challenges in resolving cell-type contributions in situ. This has spurred extensive development of deconvolution methods\, yet evaluation remains lacking. In this work\, we present a novel graph attention auto-encoder for simulating spatial cell type distributions from three major tissue types including brain\, cancer and organs\, and performed a comprehensive benchmarking. \nBio: Ajita Shree is a PhD student in the Department of Computer Science and Engineering at IIT Kanpur and will be joining EMBL-EBI\, UK\, as a postdoctoral researcher. Previously\, she worked for three years as a Data Scientist at GE Aerospace and GE Global Research\, and was a graduate of GE’s Global Leadership Program. She earned her M.Tech. in Computer Science and Engineering from IIT Delhi in 2017 and holds a B.Tech. in Computer Science and Engineering. URL:https://homecse.iitd.ac.in/event/deep-generative-models-for-single-cell-and-spatial-genomics-by-ajita-shree/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi CATEGORIES:Seminars END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260130T120000 DTEND;TZID=Asia/Kolkata:20260130T130000 DTSTAMP:20261010T113105 CREATED:20251227T051556Z LAST-MODIFIED:20251227T051556Z UID:2253-1769774400-1769778000@homecse.iitd.ac.in SUMMARY:Image decomposition with Fluorescence Microscopy Data by Ashesh DESCRIPTION:Venue: Bharti501 \nAbstract: Fluorescence microscopy is limited by optics\, fuorophore chemistry\, and photon exposure\, forcing trade-ofs in speed\, resolution\, and depth. In this talk\, I will discuss my PhD research that addresses these challenges. Specifcally\, my PhD research enables imaging of multiple cellular structures within a single fuorescent channel\, allowing faster imaging with less photon exposure. Technically speaking\, given a superimposed image (e.g.\, containing nucleus and tubulin)\, the objective is to predict the constituent images separately. \nThis talk focuses on my frst work\, µSplit. Early in my PhD\, we found that regular deep architectures performed best with large image patches\, but GPU memory limits hindered scalability. We thus developed µSplit\, a novel meta-architecture enabling memory-efcient use of large image context. Built on Hierarchical-VAE (HVAE) and U-Net variants\, it modifes HVAE’s ELBO loss for non-autoencoding tasks\, modifes KL loss for high-frequency details extraction\, and reformulates the encoder output for stable training. We also created a synthetic dataset to evaluate our network’s capability to extract large image context. Lastly\, we explored tiling artifacts\, analyzed two mitigation strategies\, and demonstrated the superiority of one\, both empirically and via out-of-distribution arguments. \n\nBio: Ashesh is a postdoctoral fellow at Human Technopole\, Milan\, Italy. He recently completed his PhD in Computer Science at TU Dresden\, Germany\, conducted in Florian Jug’s lab at Human Technopole’s Computational Biology Center. His doctoral research focused on image decomposition\, specifcally unmixing superimposed fluorescence microscopy images into constituent channels. With frst-author publications in top CV/ML venues such as ECCV\, ICCV\, and NeurIPS\, and a recent one accepted to Nature Methods\, his work ofers a robust solution to this challenge. His thesis earned a nomination for TU Dresden’s PhD prize nominations (pending decision)\, an €8\,700 EMBO grant for a research visit to ENS de Lyon on self-supervised fnetuning and uncertainty quantifcation\, and the Best Oral Presentation Award at the 2024 HT PhD & Postdoc Symposium. Previously\, Ashesh earned a dual B.Tech+M.Tech in Computer Science (2015) from IIT Delhi\, India. He brings over three years of industry experience as a Data Scientist and served as Research Assistant at National Taiwan University under Prof. Hsuan-Tien Lin\, initiating multiple computer vision projects\, culminating in publications. URL:https://homecse.iitd.ac.in/event/image-decomposition-with-fluorescence-microscopy-data-by-ashesh/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260129T120000 DTEND;TZID=Asia/Kolkata:20260129T130000 DTSTAMP:20261010T113105 CREATED:20260124T095221Z LAST-MODIFIED:20260124T123516Z UID:2392-1769688000-1769691600@homecse.iitd.ac.in SUMMARY:Lumos: A DSL for Language Model System Certification by Isha Chaudhary DESCRIPTION:Venue: Bharti 501\nAbstract: As Language Model Systems (LMS) are deployed across an expanding range of applications\, aligning them with human ethics has become crucial. Although recent works propose methods to formally certify LMS properties such as fairness\, correct question answering\, and safety\, these approaches are generally ad hoc and hard to generalize. We introduce a principled alternative: a domain-specific language\, Lumos\, for specifying and formally certifying LMS behaviors. Lumos is the first imperative probabilistic programming language over graphs\, with constructs to generate independent and identically distributed prompts for LMS. It offers a structured view of prompt distributions via graphs\, forming random prompts from sampled subgraphs. Lumos supports certifying LMS for arbitrary prompt distributions via integration with statistical certifiers. Lumos can encode existing LMS specifications\, including complex relational and temporal specifications. It also facilitates specifying new properties – we present the first safety specifications for vision-language models (VLMs) in autonomous driving scenarios developed with Lumos. Using these\, we show that the state-of-the-art VLM Qwen-VL exhibits critical safety failures\, producing incorrect and unsafe responses with at least 90% probability in right-turn scenarios under rainy driving conditions\, revealing substantial safety risks. We further demonstrate that specification programs written in Lumos enable finding specific failure cases exhibited by state-of-the-art LMS. Lumos is the first systematic and extensible language-based framework for specifying and certifying LMS behaviors\, paving the way for a wider adoption of LMS certification.\n  \nBio: Isha Chaudhary is a 4th year Computer Science Ph.D. student at the University of Illinois Urbana-Champaign\, advised by Prof. Gagandeep Singh. Her research focuses on formal methods to make frontier models trustworthy. Her work has been recognized at top-tier conferences including ICLR 2025\, AISTATS 2026 (spotlight)\, and MLSys 2024. She earned a B.Tech. in Electrical Engineering from IIT Delhi\, where she was awarded the Institute Silver Medal and Prof. C.S. Jha Memorial Award. She earned an MS in CS from UIUC. URL:https://homecse.iitd.ac.in/event/lumos-a-dsl-for-language-model-system-certification-by-isha-chaudhary/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi ORGANIZER;CN="Subodh Sharma":MAILTO:svs@cse.iitd.ac.in END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260122T120000 DTEND;TZID=Asia/Kolkata:20260122T130000 DTSTAMP:20261010T113105 CREATED:20260116T114035Z LAST-MODIFIED:20260116T144924Z UID:2385-1769083200-1769086800@homecse.iitd.ac.in SUMMARY:Logical Relations for Formally Verified Authenticated Data Structures by Chaitanya Agarwal DESCRIPTION:Venue: Bharti501 \nAbstract: Authenticated data structures (ADSs) allow untrusted third parties to carry out operations which produce proofs that can be used to verify an operation’s output. Such data structures are challenging to develop and implement correctly. In this talk\, I will talk about a library\, Authentikit\, that is implemented in OCaml\, that generates authenticated versions of data structures automatically. I will also talk about recent work by us (https://dl.acm.org/doi/abs/10.1145/3719027.3744801) that gives a formal proof of security and correctness of Authentikit. The proof is based on a new relational separation logic for reasoning about programs that use collision-resistant cryptographic hash functions. This logic provides a basis for constructing two semantic models of a type system\, which are used to justify how Authentikit makes use of type abstraction to enforce security and correctness. Using these models we also prove the correctness of several optimizations to Authentikit and then show how optimized\, hand-written implementations of authenticated data structures can be soundly linked with automatically generated code. All of the results have been mechanized in the Rocq prover using the Iris framework. \nSpeaker Bio: Chaitanya Agarwal (https://culechetoo.github.io ) is a 3rd year computer science PhD student at the New York University\, advised by Joseph Tassarotti. He is broadly interested in programming languages and formal verification with a particular focus on verification of security applications. In the past\, he has also worked with Thomas Wies on abstract-interpretation analysis for recursive\, higher-order programs\, and with Shibashis Guha\, on developing statistical-model-checking techniques for Markov Decision Processes (MDPs). Chaitanya obtained his B.Tech. from IIIT Delhi. URL:https://homecse.iitd.ac.in/event/logical-relations-for-formally-verified-authenticated-data-structures-by-chaitanya-agarwal/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi CATEGORIES:Seminars END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260116T120000 DTEND;TZID=Asia/Kolkata:20260116T130000 DTSTAMP:20261010T113105 CREATED:20260113T105922Z LAST-MODIFIED:20260113T105922Z UID:2354-1768564800-1768568400@homecse.iitd.ac.in SUMMARY:WhiteLie: A Robust System for Spoofing User Data in Android Platforms by Harish Yadav DESCRIPTION:Venue: SIT113 \nAbstract: The Android operating system uses a permission framework that allows users to control access to their private data\, such as location and contacts\, when using apps. However\, many apps become non-functional or crash if denied these permissions\, effectively pressuring users to grant access and compromising their privacy. In this paper\, we introduce WhiteLie\, a robust user data spoofing system designed to protect user privacy by feeding spoofed data to apps without requiring device rooting or binary modification. Through experiments on 70 pre-installed and user-installed Android apps\, we demonstrate that WhiteLie successfully spoofs 78.32% of the requested permissions without detection or crashes. Unlike previous methods that involve modifying the Android OS or rebuilding app binaries\, WhiteLie operates on non-rooted devices\, maintaining full app functionality. Furthermore\, WhiteLie is able to bypass continuous authentication mechanisms\, which rely on sensor data for ongoing user validation\, highlighting critical weaknesses in such security frameworks. Our findings show that WhiteLie effectively mitigates privacy risks from malicious apps\, as demonstrated in case studies where it prevented unauthorized data uploads and reduced the success rate of side-channel attacks. Additionally\, WhiteLie enhances user control over data privacy in everyday apps like Facebook\, where it was used to detect and block unauthorized audio recordings. Despite its powerful capabilities\, WhiteLie introduces minimal performance overhead\, with only a 2.52% increase in battery consumption and negligible impact on app execution time. WhiteLie proves to be a practical and efficient solution for enhancing user privacy in the Android ecosystem\, offering users greater control over their data while ensuring seamless app functionality. \nBio: Harish Yadav(2021CSY7544) is a MSR Student in the Department Of Computer Science & Engineering\, IIT Delhi. \nHe will be defending his thesis before a panel of Professors. URL:https://homecse.iitd.ac.in/event/whitelie-a-robust-system-for-spoofing-user-data-in-android-platforms-by-harish-yadav/ LOCATION:SIT 113\, Amar Nath and Shashi Khosla School of Information Technology\, Indian Institute of Technology\, Delhi\, Hauz Khas\, New Delhi\, Delhi\, 110016\, India END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260115T110000 DTEND;TZID=Asia/Kolkata:20260115T120000 DTSTAMP:20261010T113105 CREATED:20260114T064025Z LAST-MODIFIED:20260114T064025Z UID:2361-1768474800-1768478400@homecse.iitd.ac.in SUMMARY:Agentic Information Seeking for Knowledge Acquisition by Revanth Reddy DESCRIPTION:Venue: SIT001 \nAbstract:  The vast expansion of online information has shifted the challenge from simply locating data to efficiently filtering and aggregating relevant content from diverse sources. This talk introduces innovative methodologies aimed at enhancing automated information seeking capabilities within intelligent systems. I will present a modular\, agent-based framework that decomposes the information-seeking process into navigation\, extraction\, and aggregation components. This approach enables exploratory behaviors that significantly outperform current web agents. Next\, I will discuss the application of these techniques to tackle novel challenges in knowledge acquisition across various domains: 1) Automated Wikipedia Updates: An approach to automatically update and maintain Wikipedia articles using real-time online sources\, and 2) Software Issue Localization: How enabling information-seeking over code repositories helps identify and address localization issues for software problems. \nBio: Revanth Gangi Reddy is a research scientist at Google DeepMind\, working on Gemini Post-Training. He finished his Ph.D. in Computer Science at the University of Illinois Urbana-Champaign\, advised by Prof. Heng Ji. His research interests lie in knowledge-driven natural language processing\, focusing on agentic search\, ranking models\, and retrieval-augmented generation. Revanth’s work has been published in leading conferences such as ICLR\, ACL\, AAAI\, EMNLP\, NAACL\, and SIGIR\, and he also presented a tutorial on Open-Retrieval Question Answering at IJCAI 2023. Revanth has previously done research internships at Salesforce Research\, Apple\, AI2\, IBM Research\, and Amazon Science\, and was the team lead for UIUC at the Alexa SocialBot Grand Challenge 5. Revanth holds a Bachelor’s degree in Computer Science from the Indian Institute of Technology Madras and is a Siebel Scholar (class of 2022). URL:https://homecse.iitd.ac.in/event/agentic-information-seeking-for-knowledge-acquisition-by-revanth-reddy/ LOCATION:SIT 001\, Amar Nath and Shashi Khosla School of Information Technology\, IIT Delhi\, Hauz Khas\, New Delhi 110016\, India\, Delhi\, Delhi\, 110016\, India END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260109T160000 DTEND;TZID=Asia/Kolkata:20260109T170000 DTSTAMP:20261010T113105 CREATED:20260106T111834Z LAST-MODIFIED:20260108T175420Z UID:2263-1767974400-1767978000@homecse.iitd.ac.in SUMMARY:Coping with choices - List Decoding in Coding Theory by Dr. Shashank Srivastava DESCRIPTION:Venue: Bharti-501/MS Teams \nAbstract: The goal of error correcting codes is to encode data in a way that allows for this data to be recovered even if the encoded copy is corrupted by an adversary. \nThe usual algorithmic challenge associated with codes\, called decoding\, is to output the uncorrupted copy of data by looking only at the corrupted copy. However\, when noise levels are high\, the same corrupted copy could correspond to multiple uncorrupted copies. The task of list decoding is to output all such candidates. \nIn this talk\, we will talk about what makes list decoding interesting and challenging\, and its somewhat surprising connections to other areas in CS. We will then survey recent progress in list decoding for codes based on algebra and on expander graphs. \n  \nBio: Shashank Srivastava is a joint postdoc between Institute for Advanced Study (IAS)\, Princeton and DIMACS\, Rutgers University. Before this\, he obtained a PhD in 2024 from TTI Chicago and a BTech in 2018 from IIT Kharagpur. Shashank’s research focuses on coding theory and spectral algorithms\, and his work has won Best Paper and Best Student Paper awards at SODA URL:https://homecse.iitd.ac.in/event/coping-with-choices-list-decoding-in-coding-theory-by-dr-shashank-srivastava/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260108T120000 DTEND;TZID=Asia/Kolkata:20260108T130000 DTSTAMP:20261010T113105 CREATED:20251225T164159Z LAST-MODIFIED:20251225T164159Z UID:2251-1767873600-1767877200@homecse.iitd.ac.in SUMMARY:Online Flexible Busy Time Scheduling on Heterogeneous Machines by Gruia Calinescu DESCRIPTION:Venue: Bharti501 \nAbstract: We study the online busy time scheduling model on heterogeneous machines. In our setting\, jobs with uniform length arrive online with a deadline that becomes known to the algorithm at the job’s arrival time. An algorithm has access to machines\, each with different associated capacities and costs. The goal is to schedule jobs on machines by their deadline\, so that the total cost incurred by the scheduling algorithm is minimized. While busy time scheduling has been well-studied\, relatively little is known when machines are heterogeneous (i.e.\, have different costs and capacities)\, despite this natural theoretical generalization being the most practical model for clients using cloud computing services. We make significant progress in understanding this model by designing an 8-competitive algorithm for the problem on unit-length jobs and provide a lower bound of 2 on the competitive ratio. The lower bound is tight in the setting when jobs form non-nested intervals. Our 8-competitive algorithm generalizes to one with competitive ratio 8(2p-1)/p < 16 when all jobs have uniform length p. \nJoint work with Sami Davies\, Samir Khuller\, and Shirley Zhang \n  \nBio: Gruia Calinescu has studied at University of Bucharest\, received his PhD in 1998 from Georgia Institute of Technology and has worked since 2000 at Illinois Tech. He has held short term positions at DIMACS\, U. Waterloo\, U. Wisconsin Milwaukee\, and Northwestern University\, and also visited the Max Plank Institute for Informatics and the Hausdorff Research Institute for Mathematics. \nHis best works (all of them improved or generalized by now) are on Multiway Cut\, Zero Extension\, and Maximizing a Monotone Submodular Function Subject to a Matroid Constraint. URL:https://homecse.iitd.ac.in/event/online-flexible-busy-time-scheduling-on-heterogeneous-machines-by-gruia-calinescu/ LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi END:VEVENT BEGIN:VEVENT DTSTART;TZID=Asia/Kolkata:20260105T120000 DTEND;TZID=Asia/Kolkata:20260105T130000 DTSTAMP:20261010T113105 CREATED:20260103T092244Z LAST-MODIFIED:20260103T170405Z UID:2260-1767614400-1767618000@homecse.iitd.ac.in SUMMARY:Traceable Secret Sharing: Strong Security and Efficient Constructions by Aditi Partap DESCRIPTION:Venue: Bharti 501 \nAbstract: Suppose Alice uses a t-out-of-n secret sharing to store her secret key on n servers. Her secret key is protected as long as t of them do not collude. However\, what if a less-than-t subset of the servers decides to offer the shares they have for sale? In this case\, Alice should be able to hold them accountable\, or else nothing prevents them from selling her shares. With this motivation in mind\, Goyal\, Song\, and Srinivasan (CRYPTO 21) introduced the concept of {\em traceable secret sharing}. In such schemes\, it is possible to provably trace the leaked secret shares back to the servers who leaked them. Goyal et al. presented the first construction of a traceable secret sharing scheme. However\, secret shares in their construction are quadratic in the secret size\, and their tracing algorithm is quite involved as it relies on Goldreich-Levin decoding. \nIn this work\, we put forth new definitions and practical constructions for traceable secret sharing. In our model\, some f