Learning Of Support Vector Machines With Probability Output Networks

Deep learning can extract the characteristics automatically in the learning process, and learn from unlabeled samples.Although well-trained deep networks provide good performance,the learning algorithm requires accurate configuration and user-determined hyper-parameters.The support vector machine is a shallow structure in essence, can not get as deep learning as the ability to automatically obtain features. Therefore, it is of great theoretical and practical significance to study the ability of automatically extracting the internal structure of data while maintaining the support vector machine's own advantages.In the classification model, based on the deep structure of deep learning and the SRM of support vector machine and the conditional probability estimation in probability output network,a multi-layer support vector machines (SVM) is established. Where the kernel widths were constructed as the form of a grid.The kernel parameters are obtained according to the maximum product value of the positive and negative p-values where chosen by the K-S test of the cumulative probability distribution and the empirical cumulative probability distribution of the corresponding ? distribution.The corresponding output are the extracted feature, which is used as the input of the next layer model until the model reaches the end condition. The model proves the validity of the model by three classification data sets.In the regression model, based on the deep structure of deep learning and the generalization ability of support vector machine and the conditional probability estimation in probability output network, a multi-layer support vector machines (SVM) is established. Where the kernel widths were constructed as the form of a grid.The kernel parameters are obtained according to the maximum value of the p-values where chosen by the K-S test of the cumulative probability distribution and the empirical cumulative probability distribution of the corresponding ? distribution.The corresponding output is the extracted feature, which is used as the input of the next layer model until the model reacher the end condition. The model proves the validity of the model by three standard regression data sets.