Report Title: Deep Spectral Learning: Another Path of Unsupervised Learning

Presenter: Assistant Professor Deng Zhijie

Affiliation: Shanghai Jiao Tong University

Report Time: June 9, 2023 (Friday) at 14:00

Report Location: Room 1104, Jade Science and Education Building A

Tencent Meeting ID: 567-863-534

Tencent Meeting Password: Please contact Associate Professor Hu Wenbo via email to obtain it

（http://faculty.hfut.edu.cn/huwenbo/）

Report Abstract: Unsupervised learning is a key to creating human-level intelligence, and recent widely spread models such as diffusion models and language models are representative technologies in this field. This report will discuss another path of unsupervised learning: spectral methods + deep learning. Spectral methods are a typical approach to unsupervised learning before the era of deep learning. The introduction of deep learning alleviates its low scalability problem and injects inductive bias. This report will introduce the new paradigm of deep spectral learning from three aspects: theory, algorithms, and applications, and provide an outlook on future research.

Personal Introduction: Zhi Jie Deng is an assistant professor at the Qingyuan Research Institute of Shanghai Jiao Tong University. He received his bachelor's and doctoral degrees from Tsinghua University from 2013 to 2022, with a major research focus on probabilistic modeling and deep learning. He has published more than ten academic papers in high-level conferences and journals such as ICML, NeurIPS, ICLR, CVPR, ICCV, and JMLR. He has received awards and honors such as the NVIDIA Pioneer Research Award, Outstanding Graduate of the Department of Computer Science at Tsinghua University, Tsinghua University '84 Innovation Future Award, and VALSE Focus Paper Award. He has served as a reviewer for top international conferences and journals such as ICML, NeurIPS, ICLR, CVPR, ICCV, and TIP multiple times.