Date: Friday, November 22, 2024

Time: 3:00 p.m. - 4:00 p.m.

Location: Centergy Building 5126. The associated zoom link is: https://gatech.zoom.us/j/98692371115

Speaker: Prithwijit Chowdhury

Speakers' Title: Ph.D. candidate in the Center for Energy and Geo Processing and Science at Georgia Tech

Seminar Title: Necessity and Sufficiency: A Causal Perspective on Importance

Abstract: As machine learning (ML) models are increasingly applied in critical decision-making domains, the demand for transparency and interpretability has grown significantly. Explainable AI (XAI) methods aim to shed light on model decision-making processes, offering insights into the contribution of various input features. However, the diversity of explanation strategies often leads to inconsistent and conflicting outcomes, highlighting the need for more robust and reliable evaluation frameworks. This work presents a several frameworks and algorithms rooted in the causal principles of necessity and sufficiency to assess and enhance the robustness of explanation methods across a wide range of applications. By systematically quantifying these concepts, our framework provides a unified approach to evaluating the fidelity and consistency of explanations, offering a deeper understanding of model behavior. Furthermore, we extend the applicability of this framework to reinforcement learning, leveraging its capabilities to optimize decision-making processes and enhance user interactions. By integrating our framework into these dynamic environments, we aim to provide actionable insights that improve both system performance and human understanding. This comprehensive approach establishes a robust foundation for advancing the interpretability and reliability of ML models, ensuring their safe and effective deployment in complex, high-stakes scenarios where trust, accountability, and informed decision-making are essential.

Bio: Prithwijit Chowdhury received his B.Tech. degree from KIIT University, India, in 2020. He joined the Georgia Institute of Technology as an MS student in the Department of Electrical and Computer Engineering in 2021. He is currently pursuing his Ph.D. degree as a researcher in The Center for Energy and Geo Processing (CeGP) and is a member of the Omni Lab for Intelligent Visual Engineering and Science (OLIVES). His research interests include digital signal and image processing, and machine learning, with a focus on applications in geophysics. He is an IEEE Student Member and has published his work in prominent conferences such as IMAGE 2023, IMAGE 2024, and ICIP 2024.