A study on GMLVQ convex and non-convex regularization

Abstract

In this work we investigate the effect of convex and non-convex regularization on the Generalized Matrix Learning Vector Quantization (GMLVQ) classifier, in order to obtain sparse models that guarantee a better generalization abil-ity. Three experiments are used for evaluating six different sparse models in terms of classification accuracy and qualitative sparseness. The results show that non-convex models outperform traditional convex sparse models and non-regularized GMLVQ.