Generalization theory and generalization methods for neural networks

Generalization ability is the most important performance of a feed-forward neural network, and the problem of generalization is widely studied recently among the neural network community. Research on this subject can be divided into two fields: generalization theory discusses the factors that affect the generalization ability, while generalization methods try to find algorithms for improved performance. This survey reviewed the main results on generalization research, and tried to point out the relationship between generalization theory and corresponding generalization methods. A prospect on generalization research was also given.