BEGIN:VCALENDAR X-WR-CALDESC:Yale Department of Computer Science X-WR-CALNAME:Yale CS Events BEGIN:VEVENT SUMMARY:CS Talk - Fatemeh Nargesian DTSTART:20230403T200000 DTEND:20230403T210000 DESCRIPTION:Event description:\nCS Talk\nFatemeh Nargesian\n\nHost: Anurag Khandelwal\n\nTitle: Data Lakes: Discovery\, Data Debiasing\, and Query A nswering\n\nAbstract:\n\nData for AI is increasingly reliant on the integr ation of multiple sources – sometimes obtained from open data repositori es or data marketplaces. Despite decades of research in data integration a nd cleaning\, we are still not sure how to construct AI-ready structured d atasets – data with descriptive features and representative distribution . In this talk\, first\, I will describe how to discover relevant datasets \, based on join and union operations from large-scale data repositories\, by designing efficient index structures. Next\, I will show how to tailor a dataset with desired distribution requirements from multiple sources\, in order to construct unbiased datasets. We will also see how to obtain an IID sample over normalized data\, to improve the efficiency of model trai ning and perform approximate query answering. Finally\, I will conclude by discussing distribution-aware and human-centric aspects of the management of data lakes.\n\nBio:\n\nFatemeh Nargesian is an assistant professor of computer science at the University of Rochester. She obtained her PhD at t he University of Toronto. Her research interests are in data management fo r AI-based data analytics and scientific time-series management. Her work has appeared at top-tier venues including VLDB\, SIGMOD\, and ICDE\, and h as won the best demo award of VLDB 2017.\n\n\nhttps://cpsc.yale.edu/event/ cs-talk-fatemeh-nargesian LOCATION:AKW 200 STATUS:CONFIRMED URL:https://cpsc.yale.edu/event/cs-talk-fatemeh-nargesian END:VEVENT BEGIN:VEVENT SUMMARY:CS Talk - David Held DTSTART:20230404T200000 DTEND:20230404T210000 DESCRIPTION:Event description:\nCS Talk\nDavid Held\n\nHost: Brian Scassel lati and Tesca Fitzgerald\n\nTitle: Relational Affordance Learning for R obot Manipulation\n\nAbstract:\n\nRobots today are typically confined to i nteract with rigid\, opaque objects with known object models. However\, th e objects in our daily lives are often non-rigid\, can be transparent or r eflective\, and are diverse in shape and appearance. I argue that\, to enh ance the capabilities of robots\, we should develop perception methods tha t estimate what robots need to know to interact with the world. Specifical ly\, I will present novel perception methods that estimate “relational  affordances”: task-specific geometric relationships between objects that allow a robot to determine what actions it needs to take to complete a ta sk. These estimated relational affordances can enable robots to perform complex tasks such as manipulating cloth\, articulated objects\, grasping transparent and reflective objects\, and other manipulation tasks\, gener alizing to unseen objects in a category and unseen object configurations. By reasoning about relational affordances\, we can achieve robust perfor mance on difficult robot manipulation tasks.\n\nBio:\n\nDavid Held is an a ssistant professor at Carnegie Mellon University in the Robotics Institute and is the director of the RPAD lab: Robots Perceiving And Doing. His res earch focuses on perceptual robot learning\, i.e. developing new methods a t the intersection of robot perception and planning for robots to learn to interact with novel\, perceptually challenging\, and deformable objects. Prior to coming to CMU\, David was a post-doctoral researcher at U.C. Berk eley\, and he completed his Ph.D. in Computer Science at Stanford Universi ty.  David also has a B.S. and M.S. in Mechanical Engineering at MIT.  D avid is a recipient of the Google Faculty Research Award in 2017 and the N SF CAREER Award in 2021.\n\n\nhttps://cpsc.yale.edu/event/cs-talk-david-he ld LOCATION:AKW 200 STATUS:CONFIRMED URL:https://cpsc.yale.edu/event/cs-talk-david-held END:VEVENT BEGIN:VEVENT SUMMARY:CS Talk - Andrea Zanette DTSTART:20230411T200000 DTEND:20230411T210000 DESCRIPTION:Event description:\nCS Talk\nAndrea Zanette\n\nHost: Yang Cai\ n\nTitle: Towards a Statistical Foundation for Reinforcement Learning\n\n Abstract:\n\nIn recent years\, reinforcement learning algorithms have achi eved a number of headline-grabbing empirical successes on various complex tasks. However\, applying the reinforcement learning paradigm to new probl ems remains highly challenging. In many cases\, the existing algorithms ne ed to be modified\, and new ones may have to be developed to solve the pro blem at hand. In order to do so effectively\, we must gain some understand ing about the foundations of reinforcement learning.\n\nIn this talk I wil l present some recent results of my research towards this goal. I will fir st present an algorithm that can exploit the domain structure to learn muc h faster on easier problems\, while retaining state-of-the art worst-case guarantees on pathologically hard ones.  Then I will discuss a fundamenta l information-theoretic lower-bound\, which establishes that reinforcement learning can be exponentially harder than supervised learning even when s imple linear predictors are implemented. Finally\, I will discuss a stati stically optimal algorithm to learn from historical data.\n\nBio:\n\nAndr ea Zanette is a postdoctoral scholar in the Department of Electrical Engin eering and Computer Sciences at the University of California\, Berkeley\, supported by a fellowship from the Foundation of Data Science Institute. He completed his PhD (2017-2021) in the Institute for Computational and Ma thematical Engineering at Stanford University\, advised by Prof Emma Bruns kill and Mykel J. Kochenderfer. His PhD dissertation investigated modern Reinforcement Learning challenges such as exploration\, function approxima tion\, adaptivity\, and learning from offline data. His work was supported by a Total Innovation Fellowship and his PhD thesis was awarded the Gene Golub Outstanding Dissertation Award from his department. Andrea’s back ground is in mechanical engineering. Before Stanford\, he worked as a soft ware developer in high-performance computing\, as well as at the von Karma n Institute for Fluid Dynamics\, a NATO-affiliated international research establishment.\n\n\nhttps://cpsc.yale.edu/event/cs-talk-andrea-zanette LOCATION:AKW 200 STATUS:CONFIRMED URL:https://cpsc.yale.edu/event/cs-talk-andrea-zanette END:VEVENT BEGIN:VEVENT SUMMARY:CS/Econ Talk - Aaron Roth DTSTART:20230414T143000 DTEND:20230414T153000 DESCRIPTION:Event description:\nCS/Econ Talk\nAaron Roth\n\nHost: Yang Cai \n\nTitle: Robust and Equitable Uncertainty Estimation\n\nAbstract:\n\nMac hine learning provides us with an amazing set of tools to make predictions \, but how much should we trust particular predictions? To answer this\, w e need a way of estimating the confidence we should have in particular pre dictions of black-box models. Standard tools for doing this give guarantee s that are averages over predictions. For instance\, in a medical applicat ion\, such tools might paper over poor performance on one medically releva nt demographic group if it is made up for by higher performance on another group. Standard methods also depend on the data distribution being static — in other words\, the future should be like the past.\n\nIn this talk\ , I will describe new techniques to address both these problems: a way to produce prediction sets for arbitrary black-box prediction methods that ha ve correct empirical coverage even when the data distribution might change in arbitrary\, unanticipated ways and such that we have correct coverage even when we zoom in to focus on demographic groups that can be arbitrary and intersecting. When we just want correct group-wise coverage and are wi lling to assume that the future will look like the past\, our algorithms a re especially simple. We draw on ideas and techniques from the economic li teratures on calibration\, expert testing\, and property elicitation\, as well as recent work in algorithmic fairness.\n\nThis talk is based on four papers that are joint works with Osbert Bastani\, Varun Gupta\, Changhwa Lee\, Chris Jung\, Mallesh Pai\, Georgy Noarov\, Ramya Ramalingam\, and Ra kesh Vohra.\n\nBio:\n\nAaron Roth is the Henry Salvatori Professor of Comp uter and Cognitive Science\, in the Computer and Information Sciences depa rtment at the University of Pennsylvania\, with a secondary appointment in the Wharton statistics department. He is affiliated with the Warren Cente r for Network and Data Science\, and co-director of the Networked and Soci al Systems Engineering (NETS) program.  He is also an Amazon Scholar at A mazon AWS. He is the recipient of a Presidential Early Career Award for Sc ientists and Engineers (PECASE) awarded by President Obama in 2016\, an Al fred P. Sloan Research Fellowship\, an NSF CAREER award\, and research awa rds from Yahoo\, Amazon\, and Google.  His research focuses on the algori thmic foundations of data privacy\, algorithmic fairness\, game theory\, l earning theory\, and machine learning.  Together with Cynthia Dwork\, he is the author of the book “The Algorithmic Foundations of Differential P rivacy.” Together with Michael Kearns\, he is the author of “The Ethic al Algorithm”.\n\n\nhttps://cpsc.yale.edu/event/csecon-talk-aaron-roth LOCATION:AKW 200 STATUS:CONFIRMED URL:https://cpsc.yale.edu/event/csecon-talk-aaron-roth END:VEVENT END:VCALENDAR