Seminar: Control Theory as a Catalyst for Modern Machine Learning Development, 22nd August, 1:00pm

When: Thursday 22nd of August, 1:00pm AEST

Where: This seminar will be partially presented at the ACFR seminar area, J04 lvl 2 (Rose St Building) and partially online via Zoom. RSVP

Speaker: Dr Chi Zhang

Title: Control Theory as a Catalyst for Modern Machine Learning Development

Abstract:
While machine learning has frequently been employed in recent control applications, this presentation explores the inverse relationship: how control knowledge can enhance the development of modern machine learning. We will first examine various areas of contemporary ML research, including approximation capability and optimization methods, through the lens of control knowledge, such as controllability analysis and optimal control. Subsequently, we will focus on recent studies of Low-Rank Adaptation (LoRA), a method that has become the default approach to train today’s large language models (LLMs). By reinterpreting these LoRA modules as controllers, we offer a novel perspective and develop a set of sufficient conditions for controller design.

Bio:
Dr Zhang is a Research Scientist at National University of Singapore (NUS). Prior to this, he was a senior research scientist in Institute of High Performance Computing (IHPC), Singapore. He received his Ph.D. degree from Nanyang Technological University in 2019 and BSc degree from University of Science and Technology of China in 2014. Dr Zhang’s recent research focuses on integrating classical control theory with modern machine learning techniques, with a particular emphasis on transfer learning. He has published numerous papers in prestigious conferences and journals, including ICML, ICCV, ECCV, IJCAI, and AIJ.