Tutorial 1

From Theory to Practice: Diffusion Models in Sequential Decision-Making

9:00-10:15, Dec. 18, 2024, UTC+8. Seminar Room: Level 4, AESR 4-1

Abstract
Diffusion models have emerged as a powerful alternative to traditional generative models, offering improved sample quality and training stability. This tutorial provides an in-depth exploration of the application of diffusion models in enhancing sequential decision-making solutions, aiming to inspire new research in this rapidly evolving area. We begin by examining the core challenges that traditional sequential decision-making algorithms face, followed by a structured taxonomy of current methods that leverage diffusion models to address these challenges. Additionally, we highlight a series of successful real-world applications where diffusion models have been effectively integrated, showcasing their practical advantages across diverse domains.

Speakers Information

Tian Long
Tian Long
About the Speaker
Ting Long is an Associate Professor at Jilin University, specializing in reinforcement learning and information retrieval. With extensive research experience in generative models and reinforcement learning, she will cover the theoretical foundations of diffusion models and their taxonomy within decision-making. Ting Long will also discuss how these models address key challenges in decision-making contexts.
Details
Zhengbang Zhu
Zhengbang Zhu
About the Speaker
Zhengbang Zhu is a fourth-year PhD student at Shanghai Jiao Tong University, with research expertise in imitation learning, reinforcement learning, and robotics applications. He will introduce representative algorithms in each category, providing practical insights, recommendations, and hands-on experiences for applying diffusion models in various sequential decision-making scenarios.
Details
Weinan Zhang
Weinan Zhang
About the Speaker
Weinan Zhang is now a Professor (with tenure) at the Department of Computer Science and Engineering, Shanghai Jiao Tong University. His research interests include reinforcement learning, agents, and large models for decision-making, with various real-world applications of robotic control, game AI, recommender systems, etc. He has published over 200 research papers at prestigious international conferences and journals, accumulating over 20k citations on Google Scholar, and has been selected as an Elsevier China Highly Cited Researcher. He has been serving as an area chair at ICML, NeurIPS, ICLR, KDD, etc., and as an associate editor at TPAMI and FCS. Weinan served as a PC co-chair in DAI 2023.
Details