论文信息 - A neural network that uses evolutionary learning

A neural network that uses evolutionary learning

This paper proposes a new neural architecture (Nessy) which uses evolutionary optimization for learning. The architecture, the outline of its evolutionary algorithm and the learning laws are given. Nessy is based on several modifications of the multilayer backpropagation neural network. The neurons represent genes of evolutionary optimization, referred to as solutions. Weights represent probabilities and are used for selection. The training value of the output layer is set to zero, the theoretical limit of every cost-oriented optimization, and the crossover operator is replaced by a transduction operator. Mutation is used as usual. Nessy algorithm can be characterized as an individual evolutionary algorithm, but as a neural network too. It was designed for image processing applications. A short example is presented, where the discriminative feature of two images is successfully detected by the proposed evolutionary neural network.

[1] Yoshiki Uchikawa,et al. Design of fuzzy controllers for semi-active suspension generated through the genetic algorithm , 1995, Proceedings 1995 Second New Zealand International Two-Stream Conference on Artificial Neural Networks and Expert Systems.

[2] Markus Höhfeld,et al. Improving the Generalization Performance of Multi-Layer-Perceptrons with Population-Based Incremental Learning , 1996, PPSN.

[3] Andy J. Keane,et al. Topology Design of Feedforward Neural Networks by Genetic Algorithms , 1996, PPSN.

[4] Hans-Paul Schwefel,et al. Parallel Problem Solving from Nature — PPSN IV , 1996, Lecture Notes in Computer Science.

[5] Tatsuo Higuchi,et al. Efficient learning of NN-MLP based on individual evolutionary algorithm , 1996, Neurocomputing.

[6] Y. Uchikawa,et al. A New Approach to Genetic Based Machine Learning and an Efficient Finding of Fuzzy Rules - Proposal of Nagoya Approach - , 1994, IEEE/Nagoya-University World Wisepersons Workshop.