CHAOS IN THE NETWORKS: SIMULATION OF A BIOLOGICAL LEARNING PROCESS BY TIME EVOLUTION IN AN UNSUPERVISED NEURAL NETWORK

Abstract

This paper shows an abstract computational system with interesting properties concerning the emulation of a learning process by time evolution, using an unsupervised neural net, called "Autoreflexive Monoconnected Neural Net". We show how this particular unsupervised neural network can be used to simulate chaotic activities related to learning processes, which have been observed in many physiological experiments. We demonstrate how this chaos activity is distributed inside the net system, leading to the generation of synchronous oscillations with different amplitudes and frequencies, consistent with the coherent and rhythmic activity of large numbers of neurons in the brain. These findings suggest that such local oscillations are important for information processing. The results suggest that the learning process is composed of different phases, which are characterized by interspersed chaotic and periodic behaviors. Small perturbations by changing net parameters can be used to control chaotic activity, and to stabilize unstable periodic orbits, offering a new approach for the learning process control.