چکيده
Hopfield neural networks, introduced by John Hopfield in 1982, were the first artificial model of associative memory and the precursor of today's Deep Belief Nets, which have provided a revolution in Machine Learning in recent years. In this talk, I will introduce the notion of strong, i.e., multiply learned, or equivalently, strongly stored patterns in Hopfield networks. I show that strong patterns have a large basin of attraction and that their retrieval capacity, in the presence of simple patterns, rises proportional to the square of their multiplicity or strength. This square law of attraction, which is rigorously proved by solving the mean field equations for the stochastic Hopfield networks, enables us to use strong patterns to model cognitive and behavioral prototypes as well as attachment types in developmental psychology. Psychotherapy of an individual can then be modeled, at its most basic level, as the learning of a new strong pattern whose multiplicity or strength eventually exceeds that of the strong pathological neural pattern learned earlier in life. Finally I will explain how this neural model has motivated the development of a new and integrative psychotherapeutic method, called self-attachment, which in a number of case studies so far has shown to be more effective than current available techniques
