Stable On - Line Evolutionary Learning of NN - MLPQiangfu

1371 Stable On-Line Evolutionary Learning of NN-MLP Qiangfu Zhao Abstract| To design the nearest neighbor based multilayer perceptron (NN-MLP) e ciently, the author has proposed a non-genetic based evolutionary algorithm called the R4|rule. For o -line learning, the R4|rule can produce the smallest or nearly smallest networks with high generalization ability by iteratively performing four basic operations: recognition, remembrance, reduction and review. This algorithm, however, cannot be applied directly to on-line learning because its inherent instability, which is caused by over-reduction and over-review. To stabilize the R4|rule, this paper proposes some improvements for reduction and review. The improved reduction is more robust for on-line learning because the tness of each hidden neuron is de ned by its overall behavior in many learning cycles. The new review is more e cient because hidden neurons are adjusted in a more careful way. The performance of the improved R4| rule for on-line learning is shown by experimental results. Keywords| Non-genetic evolutionary learning, on-line learning, the R4|rule, supervised competitive learning, the nearest neighbor based multilayer perceptron