Dr.Sumit K.Jha
University of Texas-San Antonio (UTSA), US
Brief Profile:
Sumit K. Jha is a Professor of Computer Science at the University of Texas-San Antonio (UTSA). Dr. Jha received his Ph.D. in Computer Science from Carnegie Mellon University. Before joining Carnegie Mellon, he graduated with B. Tech (Honors) in Computer Science and Engineering from the Indian Institute of Technology Kharagpur. Dr. Jha has worked on R&D problems at Microsoft Research India, General Motors, INRIA France, and the Air Force Research Laboratory. His research has been supported by the National Science Foundation (NSF), Defense Advanced Research Project Agency (DARPA), the Office of Naval Research (ONR), the Air Force Office of Scientific Research (AFOSR), the Oak Ridge National Laboratory (ORNL), the Royal Bank of Canada, the Florida Center for Cybersecurity, the Air Force Research Laboratory (AFRL) and an entity that does not want to be identified. He is a full member of Sigma Xi. Dr. Jha was awarded the prestigious Air Force Young Investigator Award and his research has led to four Best Paper awards.
Title of Talk:
Trust in Artificial Intelligence
Abstract of Talk:
As artificial intelligence continues to permeate our daily lives, our ability to quantify our trust in a specific decision of an AI system becomes a matter of paramount importance. In this talk, we will briefly cover four interrelated recent results that enable trust in AI: (1) First, we will present the attribution-based confidence (ABC) metrics from our NeurIPS’19 paper and show its relevance as a measure for quantifying trust in neural network decisions (2) Second, we will discuss how neural stochastic differential equations (SDEs) lead to visually sharper and quantitatively robust attributions compared to traditional residual neural networks. (3) Building on this earlier IJCAI’21 work, we will highlight how attributions themselves can be used to shape the noise in neural SDEs and how this leads to more robust attributions (AAAI’22). (4) Finally, we will take a short detour into trust issues in AI hardware and present our flow-based computing architecture (DAC’22) that permits robust computing even on emerging memory devices like memristors.