Cooperative Recurrent Neural Network for Multiclass Support Vector Machine Learning
Binary classification problem can be reformulated as one optimization problem based on support vector machines and thus is well solved by one recurrent neural network (RNN). Multi-category classification problem in one-step method is then decomposed into two sub-optimization problems.In this paper, we first modify the sub-optimization problem about the bias so that its computation is reduced and its testing accuracy of classification is improved. We then propose a cooperative recurrent neural network (CRNN) for multiclass support vector machine learning. The proposed CRNN consists of two recurrent neural networks (RNNs) and each optimization problem is solved by one of the two RNNs. The proposed CRNN combines adaptively the two RNN models so that the global optimal solutions of the two optimization problems can be obtained. Furthermore, the convergence speed of the proposed CRNN is enhanced by a scaling technique. Computed results show the computational advantages of the proposed CRNN for multiclass SVM learning.
Unable to display preview. Download preview PDF.
- 11.Xia, Y.S., Wang, J.: A One-layer Recurrent Neural Network for Support Vector Machine Learning. IEEE Transactions on Systems, Man and Cybernetics - Part B, 1261–1269 (2004)Google Scholar