A Hebbian learning algorithm for training a neural circuit to perform context-dependent associations of stimuli

Henghui Zhu, Ioannis Ch. Paschalidis, Michael E. Hasselmo

Abstract: We propose a biologically plausible learning algorithm to train a neural circuit model to perform context-dependent associations of stimuli with correct responses. The specific cognitive task we consider requires the ability to learn a context-dependent association rule and generalize beyond what has been seen during training. We analyze the learning algorithm using a Markov chain framework and establish its convergence. Using numerical simulation, we validate the performance of the learning algorithm and the generalization ability of the neural circuit model.