LEARNING METHOD AND SPEECH RECOGNITION USING CHAOTIC RECURRENT NEURAL NETWORKS

摘要

The present invention relates to a chaotic circulatory neural network, a learning method and a voice recognition method using the same, and is composed of chaotic neurons having dynamic properties, and the dynamics of the chaotic neurons are transferred to other neurons as well as to their previous state as well as to external input. It can be determined by the state, so that the problem of local minimization can be solved when the learning process is performed by using the steepest drop method to handle the time-varying conversion between input and output, and the learning method and the voice recognition method using the chaotic cyclic neural network. This can improve learning performance and speech recognition.