New Multi-View RVFL Network Improves Classification with Graph Embedding

A new machine learning model, XGRVFL-MV, has been introduced to improve multi-view classification tasks. This model builds upon the Random Vector Functional Link (RVFL) network framework, which is known for its efficient randomized learning capabilities. The core innovation lies in its ability to effectively integrate information from diverse data views while addressing common challenges such as maintaining the unique geometric structure of each view, mitigating the impact of large prediction errors, and modeling inter-view relationships. XGRVFL-MV achieves this by constructing RVFL representations for each view and embedding graph structures using intrinsic and penalty graphs, which are weighted via Local Fisher Discriminant Analysis. A key component is the "FleXi Guardian (XG) loss," a bounded and asymmetric loss function designed for robust residual learning. Furthermore, a residual-coupling term ensures consistency among prediction residuals across different views, without compromising the distinct representations of each view. The optimization problem is solved efficiently using an inversion-free first-order method with Nesterov accelerated gradient descent. Extensive evaluations on several benchmark datasets, including UCI, KEEL, AwA, and Corel5k, demonstrate that XGRVFL-MV delivers competitive classification performance compared to existing baseline methods. The results, supported by statistical and hyperparameter sensitivity analyses, highlight the model's effectiveness in leveraging multi-view data for improved classification accuracy.