zbMATH — the first resource for mathematics

Learning multiclass pattern discrimination. (English) Zbl 0678.68090
Genetic algorithms and their applications, Proc. 1st Int. Conf., Pittsburgh/PA 1985, 74-79 (1988).
Summary: [For the entire collection see Zbl 0671.00021.]
Genetic algorithms (GA’s) are powerful, general purpose adaptive search techniques which have been use successfully in a variety of learning systems. Previous implementations have tended to use scalar feedback concerning the performance of alternate knowledge structures on the task to be learned. This approach was found to be inadequate when the task was multiclass pattern discrimination. By providing the GA with multidimensional feedback, a problem of this type was successfully learned. In addition, a careful balance of reward and punishment was found to be necessary in order to guide the opportunistic GA to a correct solution of the problem.

68T10 Pattern recognition, speech recognition
68T05 Learning and adaptive systems in artificial intelligence