The hidden layer design for staked denoising autoencoder

摘要

Deep learning can achieve the complex function approximation and the characteristics of the input data by studying a deep nonlinear network. At present, one of the most important problems in the study of deep learning is how to construct a reasonable structure. This paper studies the deep learning model of stacked denoising autoencoder (SDA) and the remaining task is to construct its reasonable model. We introduce three effective methods to construct the structure of the SDA. Numerical experiments imply that the structure obtained by the golden section principle performs the best.