CSE Faculty Candidate Seminar - Chirag Agarwal

Name: Chirag Agarwal, postdoctoral fellow at Harvard University

Date: Thursday, March 7, 2024 at 11:00 am

Location: Scheller College of Business Building, 1st Floor, Room 102 (Google Maps link)

Link: The recording of this in-person seminar will be uploaded to CSE's MediaSpace

Title: Trustworthy ML in the Era of Foundation Models

Abstract: Machine learning (ML) models have become ubiquitous in the last decade, and with their increasing use in critical applications (e.g., healthcare, financial systems, and crime forecasting), it is vital to ensure that ML developers and practitioners understand and trust their decisions. This problem has become paramount in the era of Foundation models, which are developed by training billion parameter models using broad uncurated datasets and great computing. In this talk, I will explore key connections between different Trustworthy properties, which currently exist in dedicated silos, and their implications on the training, inference, and evaluation stages of the machine learning model pipeline: First, I will discuss NIFTY, the first trustworthy training algorithm that enables learning accurate, fair, and stable representations for high-stakes applications. Second, I will describe DeAR, a novel framework to incorporate trustworthiness properties during inference in foundation models when we do not have access to model architecture or weights. Finally, I will discuss my benchmarking efforts in explainability research for structured and unstructured data.

Bio: Chirag Agarwal is a postdoctoral fellow at Harvard University. His research interests include developing trustworthy machine learning frameworks for scalable and reliable models focusing on explainability, robustness, fairness, and privacy. He has authored in top-tier machine learning and computer vision conferences and leading scientific journals. His research has received Spotlight and Oral presentations at NeurIPS, ICML, CVPR, and ICIP. His work has been used to understand industry-scale models in Amazon, and have received industrial grants from Adobe, Microsoft, and Google.