演讲人Speaker：Dr. Wel Huang

题目Title: Feature Learning Theory in Deep Learning

时间Date：2024年5月 28日       Time：上午 14:00 ~ 16:00

地点Venue： 信息楼 1417 会议室

内容摘要Abstract：

Feature learning theory has recently gained prominence as an important theoretical framework capable of tracking the dynamics of gradient descent and characterizing the generalization in deep learning models. This evolving theory offers a novel perspective for understanding and improving the algorithms that underpin modern artificial intelligence systems. In this presentation, we aim to establish a fundamental understanding of feature learning theory, setting the stage for an in - depth exploration of its practical applications across three critical areas: Graph Neural Networks (GNNs), Out - of - Distribution (OOD) generalization, and Federated Learning (FL).

个人简介(About the speaker):

Dr. Wel Huang is a Research Scientist in the Deep Learning Theory Team at RIKEN AIP, Tokyo, working with Professor Taiji Suzuki. He obtained his Ph.D. in Computer Science from the University of Technology Sydney, guided by Professor Richard Xu, and holds a Master's degree in Statistical Physics from the University of Science and Technology of China. Dr. Huang focuses on exploring the theoretical foundations of interpretability and transparency in deep learning, optimization and generalization, as well as on developing new algorithms, models, and methodologies that enhance the interpretability and improve their performance in graph neural networks, large foundation model and Auto - ML. His contributions to the field are documented in publications such as NeurIPS, ICLR, and ICML etc. Dr. Huang also keeps a comprehensive and popular blog on the latest Deep Learning Theory works on social media, with more than 10k followers.