Novel Approach for Promoting the Generalization Ability of Neural Networks

The generalization ability of Neural Networks (NNs) is an important performance criterion of NNs. Researchers of this domain have been making an effort to promote the generalization ability of NNs. For solving this problem, people had presented several methods, for example, early stopping, regularization, result-feedback, fuzzification of input vector, neural network ensembles, etc. Although these methods can improve the generalization ability of NNs to some extent, but in general, the problem of NNs' generalization is still not solved or not completely solved. Because the essential character of artificial neural networks is of instance-based learning, it is impossible that NNs can solve all the problems by learning from limited examples.