Associative storage of complex sequences in recurrent neural networks

The problem of modelling storage and associative recall of complex sequences in recurrent neural networks is defined in the context of human memory. A linear model and a learning rule based on Hebb´s principle are reviewed. Two additional rules, one based on an iterative approach and the other based on linear programming, are presented. The performances of these three rules in terms of their storage capacity and noise tolerance during recall are compared by means of numerical simulations. Using a Monte-Carlo technique, the fractional volume of tubes of attraction around stored complex sequences is computed for each rule. Enhancements to the linear model and possible directions for future work conclude the paper