Speaker: Assistant Professor Zhijie Deng

Affiliation: Shanghai Jiao Tong University

Organizer: School of Computer Science and Information Engineering

Time: 14:40, Thursday, October 16, 2025

Venue: Lecture Hall B501, Science and Education Building, Feicui Lake Campus

Report Abstract:

Multimodal generative models represented by autoregressive models and diffusion models are current cutting-edge hotspots in the field of artificial intelligence, but the two types of models have their own advantages, disadvantages and applicable scenarios. This report will explore their organic combination, mainly focusing on the idea of diffusion for AR to solve the problems of autoregressive models in continuous signal modeling, inference efficiency and other aspects, and briefly introduce the application of relevant methods in cross-modal generation, VLA and other scenarios.

Speaker Profile:

Zhijie Deng received his Bachelor's degree (2017) and Ph.D. (2022) from the Department of Computer Science and Technology, Tsinghua University, and is currently an Assistant Professor at Shanghai Jiao Tong University. His main research focuses on generative models, with representative works including D2F (the first open-source diffusion language model with faster generation speed than autoregressive models) and Orthus (one of the earliest multimodal large language models with native image generation capabilities). Relevant technologies are applied to industry large models such as Meituan LongCat and NextStep.

He has published nearly 50 papers (more than 30 as first/corresponding author) in conferences and journals such as ICML, NeurIPS and CVPR, including many Spotlights. He serves as an Area Chair of conferences such as ICLR and CVPR, and has won awards such as the NVAIL Pioneering Research Award. He presides over a number of national/provincial and ministerial funds and CCF industry-university cooperation funds.