When is Reinforcement Learning Possible from Small Data?

Speaker: Prof. Lin Yang
Affiliation: University of California, Los Angeles

Abstract: Deep learning is often seen as the “breakthrough” AI technology of recent years, revolutionizing areas spanning computer vision, natural language processing, and game playing.  However, if we seek to deploy such systems in real-world, safety-critical domains, a starker reality emerges: deep learning systems are notoriously brittle, sensitive to so-called adversarial attacks, where an adversary manipulates inputs to the algorithm to vastly degrade its performance (both at training time or test time).  In this talk, I will present recent progress in developing new deep learning systems that are _provably_ robust against such attacks.  Specifically, I will present two paradigms for building robust deep learning architectures: convex relaxations and randomized smoothing.  I will discuss how these approaches can be used to build classifiers that are robust against test-time data manipulation and highlight recent work on using similar techniques to build classifiers that are provably secure against training-time attacks (also known as data poisoning attacks).  I will end with some discussion on the challenges that remain in robust deep learning, and the potential directions forward.

Biography: Dr. Lin Yang is an Assistant Professor of the Electrical and Computer Engineering Department at UCLA. Prior to this, he was a postdoctoral researcher at Princeton University. He obtained two Ph.D. degrees simultaneously in Computer Science and in Physics & Astronomy from Johns Hopkins University in 2017. He obtained a bachelor’s degree from Tsinghua University. His research focuses on developing fast algorithms for large-scale optimization and machine learning. His algorithms have been applied to real-world applications including accelerating astrophysical discoveries and improving network security. He has published numerous papers in top Computer Science conferences including NeurIPS, ICML, STOC, and PODS. At Johns Hopkins, he was a recipient of the Dean Robert H. Roy Fellowship.

For more information, contact Prof. Suhas Diggavi ()

Date/Time:
Date(s) - Oct 21, 2019
12:30 pm - 1:30 pm

Location:
EE-IV Shannon Room #54-134
420 Westwood Plaza - 5th Flr., Los Angeles CA 90095