BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Computer Science and Engineering - ECPv6.13.0//NONSGML v1.0//EN
CALSCALE:GREGORIAN
METHOD:PUBLISH
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:20250501T110000
DTEND;TZID=Asia/Kolkata:20250501T120000
DTSTAMP:20260924T023848
CREATED:20250425T023119Z
LAST-MODIFIED:20250427T060150Z
UID:1521-1746097200-1746100800@homecse.iitd.ac.in
SUMMARY:Certifying Large Language Models with LLMCert
DESCRIPTION:Speaker: Isha Chaudhary \nAbstract: Large Language Models (LLMs) are increasingly deployed in critical systems\, e.g.\, healthcare and finance and can produce incorrect and biased responses. These can cause huge social and economic losses to the deploying agencies and their clients. Conventional studies are\, however\, insufficient to thoroughly evaluate LLMs\, as they cannot scale to a large number of possible inputs and provide no formal guarantees. Therefore\, we develop and present the first family of LLM certification frameworks\, LLMCert\, consisting of certifiers providing formal probabilistic guarantees for desirable properties such as correct LLM reasoning and fairness on prohibitively large distributions of prompts. Our certificates are quantitative — they consist of provably high-confidence\, tight bounds on the probability of desirable LLM responses for random prompts sampled from a distribution. We design and certify novel specifications for bias and knowledge comprehension in individual certifiers – LLMCert-B (https://certifyllm.com/) and LLMCert-C (https://arxiv.org/abs/2402.15929)\, respectively. We illustrate bias certification for distributions of prompts created by applying varying prefixes drawn from a prefix distribution to a given set of prompts. We consider prefix distributions for random token sequences\, mixtures of manual jailbreaks\, and jailbreaks in the LLM’s embedding space to certify bias. We obtain non-trivial certified bounds on the probability of unbiased responses of SOTA LLMs\, exposing their vulnerabilities over distributions of prompts generated from computationally inexpensive prefix distributions. \nFor knowledge comprehension certification\, we design and use novel distributions of knowledge comprehension prompts with natural noise\, using knowledge graphs. We certify SOTA LLMs over specifications arising in precision medicine and general question-answering. We show previously undiscovered vulnerabilities of SOTA LLMs owing to natural noise in prompts. We also establish the first performance hierarchies with formal guarantees among SOTA LLMs\, pertaining to question-answering in precision medicine. \n  \nBio: Isha Chaudhary is a third-year Ph.D. candidate at the Siebel School of Computing and Data Science\, University of Illinois Urbana-Champaign\, advised by Prof. Gagandeep Singh. Her research interest is broadly in trustworthy foundation models and neural networks for computer systems. She graduated from a B.Tech. in Electrical Engineering from IIT Delhi in 2022. For details about her work\, please check out: https://ishachaudhary.web.illinois.edu/.
URL:https://homecse.iitd.ac.in/event/certifying-large-language-models-with-llmcert/
LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi
CATEGORIES:Seminars
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250506T120000
DTEND;TZID=Asia/Kolkata:20250506T130000
DTSTAMP:20260924T023848
CREATED:20250503T125437Z
LAST-MODIFIED:20250503T125437Z
UID:1572-1746532800-1746536400@homecse.iitd.ac.in
SUMMARY:Power and limitations of quantum computation and quantum cryptography by Dr. Srijita Kundu
DESCRIPTION:Title: Power and limitations of quantum computation and quantum cryptography \nSpeaker: Dr. Srijita Kundu \nAbstract: Quantum computers are approaching practical viability\, and quantum cryptography is already being deployed for secure communication. Understanding the capabilities and limitations of these technologies is crucial for their effective use. \nMy research lies at the intersection of quantum complexity theory and cryptography. I focus on proving what quantum computation can and cannot do in concrete models such as query and communication complexity. In this talk\, I will share results in both directions:\n1. I will talk about direct product theorems for quantum communication complexity\, which are a useful lower bound technique for quantum communication protocols.\n2. I will talk about quantum proofs being more powerful than classical proofs in query complexity.\nAdditionally\, I will talk about quantum protocols for novel cryptographic tasks such as certified deletion and uncloneable encryption\, whose security can be proved using the communication direct product theorems. \nShort Bio: Srijita Kundu completed her PhD at the Centre for Quantum Technologies in the National University of Singapore in 2021\, under the supervision of Prof. Rahul Jain. Since 2022\, she has been a postdoctoral researcher at the Institute for Quantum Computing in the University of Waterloo. She is interested in quantum complexity theory and cryptography.
URL:https://homecse.iitd.ac.in/event/power-and-limitations-of-quantum-computation-and-quantum-cryptography-by-dr-srijita-kundu/
LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi
CATEGORIES:Seminars
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250508T140000
DTEND;TZID=Asia/Kolkata:20250508T150000
DTSTAMP:20260924T023848
CREATED:20250507T062717Z
LAST-MODIFIED:20250507T063432Z
UID:1581-1746712800-1746716400@homecse.iitd.ac.in
SUMMARY:Trading Prophets: How to trade multiple stocks optimally
DESCRIPTION:Speaker: Surbhi Rajput\, MSR Student\, CSE Dept.\, IIT Delhi \nAbstract:\n\nIn the (single stock) \emph{trading prophet} problem formulated by Correa et\nal.\ [2023]\, an online algorithm observes a sequence of prices of a stock.\nAt each step\, the algorithm can either buy the stock by paying the current\nprice if it doesn't already hold the stock\, or it can sell the currently\nheld stock and collect the current price as a reward. The goal of the\nalgorithm is to maximize its overall profit. Correa et al.\ showed that the\noptimal competitive ratio for this problem is $\nicefrac{1}{2}$ when the\nstock prices are identically and independently distributed.\nIn this talk\, I will discuss the simplifications and generalizations of\nCorrea et al.'s analysis\, which led us to generalize the model by allowing\nthe algorithm to trade multiple stocks. First\, we generalize the model to\n$(k\,\ell\, \ell')$-\textsc{Trading Prophet Problem}\, wherein there are $k$\nstocks in the market\, and the online algorithm can hold up to $\ell$ stocks\nat any time\, where $\ell \leq k$. The online algorithm competes against an\noffline algorithm that can hold at most $\ell' \leq \ell$ stocks at any\ntime. Under the assumption that prices of different stocks are independent\,\nwe show that\, for any $\ell$\, $\ell'$\, and $k$\, the optimal competitive\nratio of $(k\,\ell\, \ell')$-\textsc{Trading Prophet Problem} is\n$\min\left\{\frac{1}{2}\,\frac{\ell}{k}\right\}$.\nWe further generalize it to $\mathcal{M}$-\textsc{Trading Prophet Problem}\nover a matroid $\mathcal{M}$ on the set of $k$ stocks\, wherein the stock\nprices at any given time are possibly correlated (but are independent across\ntime). The algorithm is allowed to hold only a feasible subset of stocks at\nany time. We prove a tight bound of $\frac{1}{1+d}$ on the competitive ratio\nof the $\mathcal{M}$-\textsc{Trading Prophet Problem}\, where $d$ is the\n\textit{density} of the matroid.\nWe then consider the non-i.i.d.\ random order setting over a matroid\,\nwherein stock prices drawn independently from $n$ potentially different\ndistributions are presented in a uniformly random order. In this setting\, we\nachieve a competitive ratio of at least $\frac{1}{1+d} - \mathcal{O}\n\left(\frac{1}{n} \right)$\, where $d$ is the density of the matroid\,\nmatching the hardness result for i.i.d.\ instances as $n$ approaches\n$\infty$.\nOur analysis of the above problems is based on the following key insights.\nFirst\, any algorithm can be simulated by one that\, on each time step\, sells\n\emph{all} its currently held stocks before buying a suitable subset of\nstocks. Second\, we prove that the general problem reduces to a restriction\nwhere the expected price of every stock is zero.\nThird\, we reduce the problem in the random order non-i.i.d.\ setting to the\ni.i.d. setting by leveraging the fact that the outcome of sampling two\nobjects without replacement from a large set is almost identically\ndistributed as the outcome of sampling with replacement.
URL:https://homecse.iitd.ac.in/event/trading-prophets-how-to-trade-multiple-stocks-optimally/
LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi
CATEGORIES:Seminars
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250514T100000
DTEND;TZID=Asia/Kolkata:20250514T110000
DTSTAMP:20260924T023848
CREATED:20250512T144512Z
LAST-MODIFIED:20250512T144600Z
UID:1589-1747216800-1747220400@homecse.iitd.ac.in
SUMMARY:Synthesis and Arithmetic of Quantum Circuits
DESCRIPTION:Speaker: Amolak Kalra (https://sites.google.com/view/amolakratankalra/home) \nAbstract: Efficient decomposition of a unitary operator U using words from a universal gate set G is a fundamental problem in quantum computing. The process by which this is achieved is called circuit synthesis. This problem arises naturally in the context of quantum circuit\ncompilation. In this talk\, I will introduce this problem and explain how one can use tools from number theory to solve it. I will then explain some recent results that build on this connection.
URL:https://homecse.iitd.ac.in/event/synthesis-and-arithmetic-of-quantum-circuits/
LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi
CATEGORIES:Seminars
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250523T120000
DTEND;TZID=Asia/Kolkata:20250523T130000
DTSTAMP:20260924T023848
CREATED:20250519T054519Z
LAST-MODIFIED:20250520T083950Z
UID:1599-1748001600-1748005200@homecse.iitd.ac.in
SUMMARY:Next-Generation AI-Enhanced Stream Processing
DESCRIPTION:Speaker: Dr. Manisha Luthra Agnihotri is the Deputy Head of the German Research Center for Artificial Intelligence (DFKI) in Darmstadt. \nIt is an online talk. Please write to the CSE office to get the Teams link. \nAbstract: In this talk\, I will outline my vision for next-generation\, AI-enhanced data management systems through the lens of learned stream processing. Today’s stream processing platforms demand extensive manual tuning to optimize critical decisions such as query plan selection\, operator placement\, and parallelism. My vision eliminates these labor-intensive processes by leveraging zero-shot learning to automatically derive optimal configurations\, thereby radically enhancing performance and generalisability. A key contribution of my work is a novel learned operator placement optimization provided by a novel cost model that forecasts the execution costs of streaming queries on heterogeneous hardware. Particularly in IoT environments—where diverse hardware and network conditions are the norm—our approach employs graph neural networks to predict query costs accurately\, even for unseen placements and query patterns. This approach not only overcomes the generalizability limitations of existing methods but also paves the way for more robust and adaptive cost-based optimizations for stream processing systems. I will also discuss my future research directions\, focusing on extending these AI-driven techniques to multi-modal stream processing. This work aims to redefine data management by creating systems that adapt to evolving computational needs for multiple modalities\, ultimately setting new standards for understanding data inputs and autonomy in stream processing. \nBio: Manisha Luthra Agnihotri is the Deputy Head of the German Research Center for Artificial Intelligence (DFKI) in Darmstadt and a Research Group Leader at TU Darmstadt. She co-leads the Systems AI for Decision Support group with focus of research on learned system optimizations and multimodal data management. Her work sits at the dynamic intersection of machine learning\, data systems\, and hardware\, with major contributions in learned cost-based optimization and the acceleration of query workloads via GPU and RDMA technologies. \nThroughout her academic journey\, Manisha has received several prestigious awards\, including the German national Best Ph.D. Thesis award from the GI/ITG special interest group on Communication and Distributed Systems (KuVS)\, the Athena Young Investigator Award\, the Anita Borg Faculty Scholarship\, the Zeiss Top Dissertation Scholarship\, and mentoring and networking accolades from the German Research Foundation (DFG). Her expertise has led her to speak at top-tier institutions such as the University of Toronto\, and she has presented her innovative research at premier conferences like SIGMOD\, VLDB\, ICDE\, and EDBT. Manisha also actively contributes to the academic community as a program committee member for major data management conferences\, including VLDB\, SIGMOD\, and EuroSys.
URL:https://homecse.iitd.ac.in/event/next-generation-ai-enhanced-stream-processing/
CATEGORIES:Seminars
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250528T120000
DTEND;TZID=Asia/Kolkata:20250528T130000
DTSTAMP:20260924T023848
CREATED:20250518T062803Z
LAST-MODIFIED:20250526T171039Z
UID:1592-1748433600-1748437200@homecse.iitd.ac.in
SUMMARY:Enhancing Safety and Ethical Alignment in Large Language Models by Rima Hazra
DESCRIPTION:Speaker:  Dr. Rima Hazra \n\n\nAbstract: In this talk\, we explore cutting-edge strategies for enhancing the safety and ethical alignment of large language models (LLMs). The research spans various approaches\, including red teaming and jailbreaking techniques\, which assess and improve model robustness and ethical integrity. We delve into how instruction-centric responses\, when generated by LLMs\, can increase the likelihood of unethical output\, thereby highlighting the vulnerabilities of these AI systems. Through the introduction of frameworks like ‘Safety Arithmetic’ and ‘SafeInfer\,’ we demonstrate methods to mitigate risks by manipulating model parameters and decoding-time behaviors to foster safer interactions. The discussions also emphasize the importance of safety alignment strategies and the challenges posed by integrating new knowledge through model edits\, which can paradoxically destabilize ethical guidelines. This comprehensive examination not only sheds light on the current vulnerabilities of LLMs but also presents a pathway toward more reliable and ethically aligned AI implementations. \n\nBio: Dr. Rima Hazra is a senior postdoc at Eindhoven University of Technology (TU\e)\, Netherlands. Earlier she was a Postdoctoral Researcher at the Singapore University of Technology and Design\, working in the area of AI safety alignment\, natural language processing\, and LLM reasoning. She earned her Ph.D. from the Indian Institute of Technology\, Kharagpur\, where she explored the area of Information retrieval\, NLP and graph learning. With experience in information retrieval\, NLP and graph learning\, Dr. Hazra has published several papers in prestigious CORE A* and A conferences such as AAAI\, ACL\, EMNLP\, NAACL\, ECIR\, ECMLP PKDD and JCDL. She has also received the prestigious Microsoft Academic Partnership Grant (MAPG) and the PaliGemma Academic Program award from Google for her work in AI safety alignment.
URL:https://homecse.iitd.ac.in/event/enhancing-safety-and-ethical-alignment-in-large-language-models-by-dr-rima-hazra/
LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi
CATEGORIES:Seminars
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Kolkata:20250529T120000
DTEND;TZID=Asia/Kolkata:20250529T130000
DTSTAMP:20260924T023848
CREATED:20250526T171008Z
LAST-MODIFIED:20250526T171008Z
UID:1628-1748520000-1748523600@homecse.iitd.ac.in
SUMMARY:Giving Some Space Can Be Hard: Two New Models to Match Agents with Locations by Shivika Narang
DESCRIPTION:Title: Giving Some Space Can Be Hard: Two New Models to Match Agents with Locations \nSpeaker: Shivika Narang (UNSW Sydney)\n\nAbstract: There can be a multitude of reasons to match agents to specific locations in a given space. In this talk\, we cover two: distributing delivery orders and assigning office spaces. For both settings\, we shall try to find solutions that satisfy desirable properties and characterize instances for which they exist.\n\nWe first initiate the study of fair distribution of delivery tasks among a set of agents\, wherein delivery jobs are placed along the vertices of a graph. Our goal is to fairly distribute delivery costs (modeled as a submodular function) among a fixed set of agents while satisfying some desirable notions of economic efficiency. We characterize instances that admit fair and efficient solutions by exploiting underlying graph structures. Unfortunately\, finding these solutions proves to be NP-hard. We complement this by designing an XP algorithm (parameterized by the number of agents) that can find all fair and efficient solutions when they exist. We conclude this discussion by theoretically and experimentally analyzing the price of fairness.\n\nWe shall then introduce and analyze distance preservation games (DPGs). In DPGs\, agents express ideal distances to other agents and need to choose locations in the unit interval while preserving their ideal distances as closely as possible. We analyze the existence and computation of location profiles that are jump stable (i.e.\, no agent can benefit by moving to another location) or welfare optimal for DPGs\, respectively.\n\nJoint Work with Hadi Hosseini and Tomasz Wąs (Fair Delivery) and Haris Aziz\, Hau Chan\, Patrick Lederer\, and Toby Walsh (DPGs).\n\nSpeaker Bio: Shivika Narang is a postdoctoral fellow at UNSW Sydney. Previously\, she was a postdoc at Simons Laufer Mathematical Sciences Institute\, Berkeley (SLMath)\, and completed her PhD from IISc Bengaluru. Her work is currently focused on fairness and efficiency in computational social choice\, especially matching and allocation problems.
URL:https://homecse.iitd.ac.in/event/giving-some-space-can-be-hard-two-new-models-to-match-agents-with-locations-by-shivika-narang/
LOCATION:Bharti 501\, IIT Campus\, Hauz Khas\, New Delhi
CATEGORIES:Seminars
END:VEVENT
END:VCALENDAR