Date：2025/7/9 14:00-15:10
Location：R107, CSIE
Speakers：Prof. Hung-Hsu Chou (Pittsburgh University)
Host：李彥寰教授
Abstract:
Despite their recent successes in various tasks, most modern machine learning algorithms lack theoretical guarantees, which are crucial to further development towards delicate tasks. One mysterious phenomenon is that, among infinitely many possible ways to fit data, the algorithms always find the "good" ones, even when the definition of "good" is not specified by the designers. In this talk I will cover the empirical and theoretical study of the connection between the good solutions in neural networks and the sparse solutions in compressed sensing, and how we can take advantage of such characteristics to perform out-of-distribution detection that is related to safety in AI. The key concepts are implicit regularization, neural tangent kernel, neural collapse, and out-of-distribution detection.
Biography:
Hung-Hsu Chou is a assistant professor in mathematics at University of Pittsburgh. My main interests are machine learning, optimization, signal processing. I am currently focusing on implicit bias/regularization, neural tangent kernel, edge of stability, conformal prediction, and adversarial attacks.