Evolution of locally defined learning rules and their coordination in feedforward neural networks

Abstract

We propose a method to evolve adaptive behavior of learning in an artificial neural network (ANN). The adaptive behavior of learning emerges from the coordination of learning rules. Each learning rule is defined as a function of local information of a corresponding neuron only and modifies the connective strength between the neuron and its neighbors. It is exposed to selective pressure based on the fitness value, which represents the importance in producing the correct output signals of the ANN. The learning rules with lower fitness values are replaced by new ones generated by genetic programming techniques. Experimental results demonstrate that the proposed method produces adaptive behavior of learning in single-layered and two-layered ANNs. This means that efficient learning rules evolve. Further, the learning rules in the two-layered ANN coordinate with each other and macroscopic adaptive behavior of learning emerges.