Training weighted associative memories by global minimization

A training strategy in a weighted associative memory (WAM) by means of global minimization is presented. The WAM emphasizes the association among patterns by imposing weights on patterns. A cost function that gives a quantitative measure of the goodness of the WAM is derived to convert the problem of finding the weights into a global minimisation, which can be solved by a gradient descent algorithm. The authors investigate the existence of the weights, prove the convergence of the training strategy, and discuss the asymptotic stability of each desired pattern and its domain of attraction. Experimental results are described