BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Lalitha Sankar (Arizona State University) DTSTART:20250224T223000Z DTEND:20250224T234500Z DTSTAMP:20260506T225204Z UID:MathandDemoc/24 DESCRIPTION:Title: Understanding Last Layer Retraining Methods for Fair Classification: Theory and Algorithms\nby Lalitha Sankar (Arizona State University) a s part of NYU CDS Math and Democracy Seminar\n\n\nAbstract\nLast-layer ret raining (LLR) methods have emerged as an efficient framework for ensuring fairness and robustness in deep models. In this talk\, we present an overv iew of existing methods and provide theoretical guarantees for several pro minent methods. Under the threat of label noise\, either in the class or d omain annotations\, we show that these naive methods fail. To address thes e issues\, we present a new robust LLR method in the framework of two-stag e corrections and demonstrate that it achieves state-of-the-art performanc e under domain label noise with minimal data overhead. Finally\, we demons trate that class label noise causes catastrophic failures even with robust two-stage methods\, and propose a drop-in label correction which outperfo rms existing methods with very low computational and data cost.\n\nLalitha Sankar is a Professor in the School of Electrical\, Computer and Energy E ngineering at Arizona State University. She joined ASU as an assistant pro fessor in fall of 2012\, and was an associate professor from 2018-2023. Sh e received a bachelor's degree from the Indian Institute of Technology\, Bombay\, a master's degree from the University of Maryland\, and a doctora te from Rutgers University in 2007. Following her doctorate\, Sankar was a recipient of a three-year Science and Technology Teaching Postdoctoral F ellowship from the Council on Science and Technology at Princeton Universi ty\, following which she was an associate research scholar at Princeton. P rior to her doctoral studies\, she was a senior member of technical staff at AT&T Shannon Laboratories.\n\nSankar's research interests are at the in tersection of information and data sciences including a background in sign al processing\, learning theory\, and control theory with applications to the design of machine learning algorithms with algorithmic fairness\, priv acy\, and robustness guarantees. Her research also applies such methods to complex networks including the electric power grid and healthcare systems . \n\nFor her doctoral work\, she received the 2007-2008 Electrical Engine ering Academic Achievement Award from Rutgers University. She received the IEEE Globecom 2011 Best Paper Award for her work on privacy of side-infor mation in multi-user data systems. She was awarded the National Science Fo undation CAREER award in 2014 for her project on privacy-guaranteed distri buted interactions in critical infrastructure networks such as the Smart G rid. She has led an NSF Institute on Data-intensive Research in Science an d Engineering (I-DIRSE)\, is a recipient of an NSF SCALE MoDL (Mathematics of Deep Learning) grant\, and a Google AI for Social Good grant. Sankar w as a distinguished lecturer for the IEEE Information Theory Society from 2 020-2022. She serves as an Associate Editor for the IEEE Transactions on I nformation Forensics and Security\, IEEE Information Theory Transactions\, and was an AE for the IEEE BITS Magazine until August 2024.\n LOCATION:https://researchseminars.org/talk/MathandDemoc/24/ END:VEVENT END:VCALENDAR