Robust pattern recognition using chaotic dynamics in Attractor Recurrent Neural Network

Strong abilities of brain, in robust and intelligent processing of data are considered in many researches. Furthermore, chaotic behavior is reported both in microscopic scale (neurons) and macroscopic one (brain behavior). Such evidences made us to incorporate chaotic behavior in artificial neural networks to increase their performance in data processing. Based on this fact, a novel chaotic Attractor Recurrent neural network (CARNN) is presented in this paper. CARNN uses chaotic nodes with quasi logistic map as activation function to create various variability around the formed attractors and a Attractor Recurrent Neural Network (ARNN) as supervisor model for evolution of these chaotic nodes to a appropriate findings. Chaotic behavior of neurons made CARNN to search effectively in attractor basins. Therefore, as results show, this model has a better performance in comparison to ARNN and Feedforward Neural Network (FNN) in robust noisy pattern recognition.