Darwinian inheritance genetic learning method of neural networks under dynamic environments

Neural network and genetic algorithms are widely known as their superior adaptation capability by imitating mechanisms of a living thing. In this paper, we proposed the Darwinian inheritance genetic learning method, where each neural network is regarded as an individual learning ability, and genetic algorithms are applied as the evolutionary processes in the population of such an individual. Especially, even if the dataset of teaching data is changed, this proposed method can find a good individual, which includes the network structures, the connection weights, and the learning parameters without starting to learn the new data set. In this paper, although the given training data set is subset of all training data, we show that our proposed method has the good performance of classification for all training data sets.