Training approaches in neural enhancement for multiobjective optimization

In previous work, a neural network was used to increase the number of solutions found by an evolutionary multiobjective optimization algorithm. In this paper, various approaches are applied in the training of the neural network to determine whether an approach exists that can provide reasonable results in a reasonable time. To this end, two heuristic training algorithms are developed. When evaluated on a suite of ten benchmark mutliobjective optimization problems, these heuristic techniques perform very well and, on average, produce many more solutions than the evolutionary multiobjective optimization approach alone. Categories and Subject Descriptors

• Type: Conference
• Year: 2008
• Where: ACMSE
• Authors: Aaron Garrett, Gerry V. Dozier
• Keywords: ACMSE 2008, Evolutionary Multiobjective Optimization, Multiobjective Optimization Algorithm, Neural Network, Theoretical Computer Science