概要:

The learning ability and the memory ability of the human brain have the excellent processing performance that the present computers can't achieve. 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.

In my laboratory, the "delayed learning method" was newly devised. Using this method, the neural network with internal memory can learn complicated time-series. However, it is not really known about this neural network's performance. Then, the performance is evaluated as compared with a neural network with feedback. A sine wave, a triangular wave, a saw tooth wave and the pulse wave were used for the time-series.

As a result, performance of the neural network with feedback went up when there is many data used for learning. On the other hand, it turned out that the mean error of learning is not greatly related to delay time on the neural network with internal memory. But, it became clear that the success rate of study was changed by delay time. The further experiment showed that the result of learning would be easy to diverge if delay time is large. As a conclusion, if delay time is little, the operation is stable. while I confirmed that study is early successful when delay time is large.