| Hi-index | 0.00 |
Artificial neural network (ANN) is a thoroughly interdisciplinary area, covering neurosciences, physics, mathematics, economics and electronics. The applications of ANN are very diverse and effective. Some drawbacks, however, have been found accompanying with the applications of ANN. In order to overcome these drawbacks, many methods have been proposed. In this article, two issues will be referred, namely models adjustment and generalization capability of ANN. Models adjustment includes two aspects: the model's parameters adjustment and the model's architecture adjustment. The purpose of the former is to improve training speed, enhance convergence and stability of network. And the purpose of the latter is to enhance recognition ability. The model's architecture is adjusted through adding a binary-coding layer to it. In order to promote the generalization capability, the perfect training sample is put forward based on mathematics.