Self-centralized jointly sparse maximum margin criterion for robust dimensionality reduction

Liangchen Hu, Jingke Xu, Lei Tian, Wensheng Zhang

Abstract: Highlights•To alleviate the deviation caused by outliers, we re-formulate the weighted MMC with L1,2<math><msub is="true"><mrow is="true"><mi is="true">L</mi></mrow><mrow is="true"><mn is="true">1</mn><mo is="true">,</mo><mn is="true">2</mn></mrow></msub></math>-norm, and update the trusted global and intra-class centroids adaptively during the iterative solving process.•The L2,1<math><msub is="true"><mrow is="true"><mi is="true">L</mi></mrow><mrow is="true"><mn is="true">2</mn><mo is="true">,</mo><mn is="true">1</mn></mrow></msub></math>-norm sparsity is introduced into the newly formulated MMC for jointly feature selection and sparse subspace learning, which significantly reduces complexity and improves generalization for the model.•A simple and efficient iterative algorithm is derived and its convergence is proved creatively and theoretically. Besides, comparative experiments demonstrate the effectiveness of the proposed method.