Evolutionary Learning of Complex Modes of Information Processing
Stefan Waner, Harold M. Hastings · Advances in Cognitive Science · 2019
This chapter describes an evolutionary learning system as an ergodic system whose states are modes of information processing. Such a system was based on three "principles of evolutionary learning". The chapter argues that evolutionary learning is an annealing process. The method of computational annealing used in solving certain optimization problems can be viewed as intermediate between gradient methods and random search methods. Moreover, annealing dynamics can be interpolated between the dynamics of both extremes by adjusting either the potential energy surface or the parameter representing temperature. The deterministic modes in system will include the additional structure of deterministic bit masks. Stochastic bit masks will be defined using suitable convex combinations of deterministic bit masks, and learning will involve driving the system toward deterministic modes. The efficiency of annealing techniques may be increased by a suitable formulation of the problem to be solved. The chapter considers speedup methods applicable in general adaptive learning via annealing.