TRKM: Twin Restricted Kernel Machines for Classification and Regression
To address the weak generalization capability and high computational cost of Restricted Kernel Machines (RKMs) on imbalanced, high-dimensional, and large-scale data, this paper proposes the Twin-Restricted Kernel Machine (TRKM). Methodologically, TRKM leverages the Fenchel–Young inequality to construct a conjugate feature duality, reformulating the optimization into a bivariate energy minimization framework that jointly optimizes visible and hidden variables. It integrates the Restricted Boltzmann Machine (RBM) energy function, kernel methods, regularized least squares, and Fenchel duality theory. Its key innovations include the first introduction of a conjugate feature duality mechanism and a twin-model architecture, substantially enhancing model robustness and scalability. Extensive experiments on UCI and KEEL benchmark datasets demonstrate consistent superiority over state-of-the-art methods. Moreover, TRKM achieves high accuracy and strong generalization in brain-age prediction—a challenging real-world neuroimaging task—validating its practical efficacy.