Visualized response surface analysis of the internal representations of neural networks

Abstract

Neural networks have a marvelous learning capability. This capability has been successfully applied in many different fields, and the contributions keep on growing at an amazing rate today. However, neural networks are always regarded as a black box. That means the internal behaviors of neural networks are unknown. The nature of the learning capability of a neural network, like its early inspiration the human brain, still remains secret.