Multi-Objective Learning via Genetic Algorithms

Genetic algorithms (GAs) are powerful, general purpose adaptive search techniques which have been used successfully in a variety of learning systems. In the standard formulation, GAs maintain a set of alternative knowledge structures for the task to be learned, and improved knowledge structures are formed through a combination of competition and knowledge sharing among the alternative knowledge structures. In this paper, we extend the GA paradigm by allowing multidimensional feedback concerning the performance of the alternative structures. The modified GA is shown to solve a multiclass pattern discrimination task which could not be solved by the unmodified GA.