ANNs, Reinforcement Learning and Expectation

Abstract

Artificial Neural Networks (ANN) are the intelligent systems that can learn through the samples presented to them. The artificial neurons based on McCullough-Pitts model are used successfully in several applications today, but they are far away from modelling the behaviour of real neurons. The Random Neural Network (RNN) model, in which signals travel as voltage spikes rather than as fixed signal levels, represents more closely the manner in which signals are transmitted in biological neural networks. Reinforcement learning is one of the approach used in training artificial systems and closely related to instrumental conditioning in animal learnin. In this paper, reinforcement learning of RNN is considered and a training rule that considers the expectation of reward is presented.