Sparse Support Vector Machine for pattern recognition

Support Vector Machine (SVM) is one of the most famous classification techniques in the pattern recognition community. However, due to outliers in the training samples, the SVM tend to be over-trained. This means that the generalization ability of the SVM will decrease for further training. In this paper, we borrow the idea of compressive sensing/sparse representation and apply it to the SVM. Our method can achieve higher classification rates than the standard SVM due to the sparser support vectors. Experimental results conducted in this paper show that our proposed technique is feasible in practical pattern recognition applications.