Title: Collaborative Expertise: Integrating Diverse Data via Large Margin Optimization

Abstract: The integration of heterogeneous information continues to pose a significant hurdle in contemporary data analysis. Although considerable advancements have been achieved, current methods frequently overlook the intrinsic variability of object patterns across distinct semantic domains. To resolve this issue, we introduce the Cooperation of Experts (CoE) framework, a system designed to encode multi-typed data into unified heterogeneous multiplex networks. By bridging gaps in modality and connectivity, CoE offers a robust and adaptable solution for deciphering the complex structures inherent in real-world datasets. Within this architecture, specialized encoders function as domain-specific experts, each dedicated to identifying unique relational patterns within particular semantic spaces. To bolster resilience and harvest complementary insights, these experts engage in collaboration driven by a novel large margin mechanism, underpinned by a customized optimization strategy. Comprehensive theoretical evaluations confirm the framework’s viability and stability, while extensive testing on various benchmarks highlights its superior performance and wide-ranging utility. The source code is publicly accessible at https://github.com/strangeAlan/CoE.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC