A learning algorithm of recurrent neural networks for discovery of units behaving like linear systems

摘要

In our previous research a priming algorithm for a recurrent neural network was proposed which discovers units behaving like linear systems and reduces the network. However, it is difficult to apply the algorithm in case that the difference between the linear units and other units is small. This paper proposes a new learning algorithm of recurrent neural networks. which eases the choice of the linear units. Furthermore the numerical results show that the algorithm is effective to achieve the purpose.