Effect of refractoriness on learning performance of a pattern sequence
Susumu Nagatoishi, Osamu Araki · 2009
The primary purpose of this study is to reveal the effects of refractoriness on learning performance. We simulated that Elman network, which consists of chaotic neurons, learns a pattern sequence using the back-propagation algorithm. Consequently, the learning speed was accelerated about 46% compared with that of the network consisting of integrate-and-fire model neurons. In addition, we analyzed the required number of hidden neurons, asynchronous activities of hidden neurons' refractoriness. These results suggested that the refractoriness contributes to efficient encoding in the hidden layer of Elman network.