The learning ability of the human brains is superior to that of the most efficient computer at present. The neural network is a technology that reproduces the human brain artificially. The hierarchical neural network that has the internal memory layer is suitable for learning of time series. We can not understand how the time series is composed even if the neural network learns the time series.
In this study, I made the neural network learn the time series, and extracted the rules from the trained neural network. There is no effective algorithm for extracting rules from the networks that treat continuous values.
At first, I made the neural network learn a sine curve. Then, I extracted the rules from the neural network by applying the extracting algorithm for the neural network that treat binary values.
The results of this study have shown that the neural network learned a sine curve successfully and that the rules extracted from the trained neural network were simple but not exact. I considered the reason is that the extracting algorithm is not suitable for the neural network used this time.
Our future task is to extract the knowledge as simple and exact rules from the trained neural network.