Analysis of Multi-View Majority Vote Learning Algorithms Through PAC-Bayesian Bounds

Abstract

The PAC-Bayesian framework has significantly advanced our understanding of statistical learning’s generalization capabilities, particularly through majority voting methods. However, its application within multi-view learning remains underexplored. This manuscript extends PAC-Bayesian analysis to address the complexities of multi-view scenarios in majority voting. We introduce novel multi-view PAC-Bayesian bounds that incorporate Rényi divergence as a nuanced complexity measure, replacing the traditional Kullback-Leibler divergence. Furthermore, we refine our theoretical framework by advancing both first and second-order bounds, as well as the C-bound. These theoretical advancements provide a robust foundation for developing generalizable machine learning models in multi-view contexts.