A General Evolutionary/Neural Hybrid Approach to Learning Optimization Problems
Dan A. Ventura, Tony R. Martinez · 1996
This paper extends work on one such hybrid method, first presented in [12], that combines the broad, parallel search capabilities of EC with the generalization and execution speed of NN. Given a function to be learned, under certain reasonable assumptions the EC can be used to solve the problem at selected points in the input space; the survivors of the evolution become the instances in a training set for the NN, which then generalizes the optimization for the entire input space. Section two of the paper presents a generalized formal description of the problem to be solved. Section three then discusses the combination of NN with EC as a general approach to solving the problem. As a proof-ofconcept for the general approach, section four describes a specific learning optimization problem and presents empirical results from simulations run on that problem. Finally, section five presents conclusions and directions for ongoing research.