Dec 4 (Wed) @ 9:30am: "Towards Robust and Cooperative Algorithms for Reinforcement Learning and Adversarial Machine Learning," Mark Beliaev, ECE PhD Defense

Research Areas: Communications & Signal Processing, Control Systems, Computer Engineering
Research Keywords: Learning from Suboptimal Demonstrations; Adversarial Machine Learning

Abstract

Recent progress in machine learning has been fueled by increasing scale, enabling breakthroughs in domains such as image generation, natural language understanding, and decision-making. While tremendous improvements have been realized for low-risk applications like chat completion, there are fundamental challenges that need to be addressed for high-risk applications like robotics: (1) Crowdsourcing introduces suboptimal data. (2) High risk settings often involve multi-agent interaction. (3) As models scale to generalize they become more susceptible to adversaries. This talk presents two separate works that consider these challenges within the domains of Reinforcement Learning and Adversarial Machine Learning. 

The first contribution introduces two algorithms for imitation learning in suboptimal settings, demonstrating that by modeling the suboptimalities present, we can improve the learning framework. The second contribution develops a robust classification algorithm designed for sparse attacks, demonstrating that by extending our theoretical insights from idealized settings, we can combat adversarial attacks in neural network classifiers. We demonstrate our results across applications in image-classification, game-playing, and robotics. Together, these contributions advance the deployment of machine learning in real-world scenarios by considering practical problem settings, presenting novel theoretical and algorithmic insights, and validating these findings empirically.

Bio

Mark Beliaev is a PhD candidate in the Department of Electrical and Computer Engineering, under the supervision of Prof. Ramtin Pedarsani. He received his dual B.E. in Electrical Engineering and Applied Mathematics from Stony Brook University, NY, in 2017, and his M.S. in Electrical and Computer Engineering from the University of California, Santa Barbara, in 2020. His research interests include reinforcement learning, game theory, and adversarial machine learning.

Hosted By: ECE Professor Ramtin Pedarsani

Submitted By: Mark Beliaev <mbeliaev@ucsb.edu>