On December 5, the Best Paper Award at ACM Multimedia Asia 2024 was bestowed on Professor Wu Xiao’s team at the School of Computing Science and Artificial Intelligence, for their research paper entitled “TMM-CLIP: Task-guided Multi-Modal Alignment for Rehearsal-Free Class Incremental Learning”. This marks the second time that this team has won a Best Paper Award at an international conference, following their success at the International Conference on Multimedia Modeling (MMM 2021).

ACM Multimedia Asia (ACM MM Asia), sponsored by the Association for Computing Machinery (ACM) and ACM Special Interest Group on Multimedia (SIGMM), is an international flagship conference in the field of multimedia. It was established through the merger of the Pacific Rim Conference on Multimedia (PCM), founded in 2000, and the International Conference on Internet Multimedia Computing and Service (ICIMCS), founded in 2009.

The paper lists SWJTU as the first affiliated institution, with 2027 Ph.D. candidate Pan Yuankang serving as the first author and Professor Wu Xiao as the corresponding author. The research was collaboratively completed by Associate Professor Yuan Zhaoquan from SWJTU, Professor Li Zechao from Nanjing University of Science and Technology, and Researcher Xu Changsheng from the Institute of Automation, Chinese Academy of Sciences.

Illustration of the TMM-CLIP Framework

Continuous learning is currently both a hot topic and critical issue in artificial intelligence. This paper focuses on class-incremental learning based on the vision-language pre-trained large model CLIP, revealing the problem of memory decay caused by visual feature confusion during the class-incremental process and proposing a corresponding task-guided multi-modal alignment solution.