BIMSA >
Advances in Artificial Intelligence
From Diffusion Model to Autoregressive––Thoughts on the Future of the World Model
From Diffusion Model to Autoregressive––Thoughts on the Future of the World Model
Organizers
Speaker
Xinyu Xiao
Time
Friday, December 13, 2024 3:00 PM - 4:00 PM
Venue
Online
Online
Zoom 787 662 9899
(BIMSA)
Abstract
From images to videos, diffusion models are demonstrating their application value in video generation. This is due to their powerful randomness and realism, which allow them to capture subtle dynamic changes, making the generated videos more authentic. Meanwhile, autoregressive models have quickly become a research hotspot in the field of video generation because of their advantages in sequence generation. They show great potential for generating smoother and more coherent videos. Furthermore, with the enhancement of computational power and optimization of model architectures, autoregressive models are continually improving in terms of generation efficiency and quality. The speaker will delve into the current advancements in image and video generation technologies by combining their cutting-edge research work in the video generation field with classical works in the area. Additionally, based on the development of visual generation and understanding, the prospects for world models are intriguing. The speaker will also discuss the research prospects and directions of world models based on current research progress. This presentation will be conducted in Chinese.
Speaker Intro
肖鑫雨，本科毕业于北京航空航天大学，博士毕业于中科院自动化研究所。目前在工业界从事人工智能研究工作，主要研究方向是视觉理解生成，包括视觉描述，视觉检索，气象预报，视觉生成，视觉识别和检测，视觉问答，强化学习，对比学习，可解释性学习，时空数据挖掘等内容。目前发表论文20余篇。