Contractive De-Noising Auto-Encoder

De-noising auto-encoder (DAE) is an improved auto-encoder which is robust to the input by corrupting the original data first and then reconstructing the input by minimizing the error function. And contractive auto-encoder (CAE) is another kind of improved auto-encoder learning robust feature by introducing Frobenius norm of the Jacobean matrix of the learned feature with respect to the input. In this paper, we combine DAE and CAE, and propose contractive de-noising auto-encoder (CDAE), which is robust to both the original input and the learned feature. We stack CDAE to extract more abstract features and apply SVM for classification. The experiment on benchmark dataset MNIST shows that CDAE performed better than CAE and DAE