Paper from Wang Youqing's team accepted by ACM MM 2026
A paper by Prof. Wang Youqing's team at Beijing University of Chemical Technology has been accepted by ACM MM 2026, a CCF-A multimedia conference. The work proposes GLCMoE, a global-local mixture-of-experts framework with dual contrastive learning for deep multi-view clustering.
The paper is titled "Global-and-Local Collaborative Mixture-of-Experts with Dual Contrastive Learning for Deep Multi-view Clustering." Wang is the first author and a graduate supervisor. Xiao Bin, a 2023 master's student in the College of Information Science and Technology, is the second author. Associate Prof. Guo Jipeng is the corresponding author and co-supervisor. Beijing University of Chemical Technology is the first completion unit and the sole corresponding unit.
According to CVer, existing deep multi-view clustering methods overlook cross-view heterogeneity and the difficulty of semantic alignment between different views. The team proposed the Global-and-Local Collaborative Mixture-of-Experts framework, or GLCMoE, to address these issues. The framework consists of a local mixture-of-experts module, a global mixture-of-experts module and a dual contrastive learning module.
The local module uses a parameter-sharing expert pool to filter cross-view heterogeneity and a gated routing mechanism to adaptively activate and reorganize collaborative experts, preserving complementary information specific to fine-grained views. The global module uses a Transformer attention-based routing mechanism to extract cross-view consistency from a unified multi-view feature space and capture high-order cross-view interactions. The dual contrastive learning module performs deep semantic alignment between views at multiple levels: coarse-grained alignment among local view-specific representations and fine-grained alignment between local fused representations and global representations.
The study also designed a gated routing balance regularizer to alleviate expert collapse and improve the utilization of each expert. The GLCMoE method follows a strategy of splitting first, aggregating later and hierarchical contrastive learning, achieving discriminative cross-view representation fusion and semantic alignment.
ACM MM is organized by the Association for Computing Machinery. Founded in 1993, it is described as a leading international conference for multimedia research, innovation and applications and as the only Class A international academic conference in multimedia, or computer graphics and multimedia, on the CCF recommendation list. The acceptance marks another breakthrough for the college in a top international conference in computer multimedia, according to the report. The work was supported by the National Natural Science Foundation of China and the Beijing Natural Science Foundation.
Editor's Summary
ACM MM 2026 has accepted a paper from Prof. Wang Youqing's team at Beijing University of Chemical Technology on GLCMoE, a global-local mixture-of-experts framework with dual contrastive learning for deep multi-view clustering. The work addresses cross-view heterogeneity and semantic alignment, and its acceptance marks a top-conference result for the institution in multimedia. The research was supported by the National Natural Science Foundation of China and the Beijing Natural Science Foundation.