Recently, the acceptance results of ACM MM 2025, the world's top international conference on multimedia technology, were announced. A paper by the research group led by Associate Professor Ma Fei from the College of Information Science and Technology was accepted by the conference. The ACM International Conference on Multimedia (ACM MM) is a top-tier international conference in the fields of computer graphics and multimedia. Hosted by ACM, this conference has, since its first edition in 1993, become a crucial platform for academic and industrial exchanges in the field. It is also a Class A international academic conference recommended by the China Computer Federation (CCF), primarily accepting high-quality papers in areas such as multimedia analysis, retrieval, generation, understanding, and human-computer interaction. The conference will be held in Dublin, Ireland, from October 27 to 31, 2025.
The accepted paper is titled Collaborative Cloud-edge Generalized Category Discovery. This work focuses on the Cloud-edge Generalized Category Discovery (CE-GCD) task. Specifically, decentralized product users typically require automated category discovery but cannot or are unwilling to provide any annotated data. Instead, they need to realize automatic discovery and recognition of new categories from a dataset containing labeled and unlabeled instances through unsupervised learning and knowledge transfer on the client side. To address this problem, the paper proposes a GCD framework that combines energy-guided known category recognition and a multi-level contrastive learning strategy. In each client, the classifier of the base model is first used to distinguish between known and unknown categories, followed by unsupervised learning on the unknown categories. Each client transmits category information through category prototypes to support learning. Experiments on multiple datasets demonstrate that this method achieves promising results.
The first author of this paper is Lecturer Liu Yingbing from the College of Information Science and Technology. It was jointly supervised by Associate Professor Ma Fei and Professor Zhang Fan, and completed in collaboration with Concordia University in Canada. Beijing University of Chemical Technology is the first completing institution. This work was supported by projects including the National Natural Science Foundation of China.