Recurrent Neural Network learning by adaptive genetic operators: Case study: Phonemes recognition
Hanen Chihi, Najet Arous · 2012
Classical training methods for Recurrent Neural Networks (RNN) suffer from being trapped in local minimal and having a high computational time. This suggests that the problems of developing methods to determine new training algorithms should be studied. This paper describes a novel hybrid method of RNN and Genetic Algorithm (GA) for phonemes recognition. We adapt the weight and bias vectors by genetic operators. In this context, we propose a mutation operators endowed with local learning rules and to apply Parent-Centric Crossover (PCX) in order to improve recognition of networks.