基于遗传算法的进化神经网络

提出了一种基于遗传算法的前馈神经网络的自动化设计方法 (genetic m ultilayer neural network,GMNN ) ,用以同时完成对网络结构空间和权值空间的搜索。该算法利用模拟退火算法、 BP算法和小生境技术来加快算法的收敛速度 ,改善解的性能。初步实验结果表明 ,该方法的收敛速度较快 ,由此得到的神经网络的泛化能力也较好 ,能够达到根据训练样本自动优化设计多层前馈式神经网络的目的。

Evolutionary neural networks based on genetic algorithms

Genetic algorithms (GAs) were used to develop an automatic optimizing method called GMNN (Genetic Multilayer Neural Network) for feedforward multilayer neural networks. The method simultaneously searches the satisfied structure and weights of the network. Simulated annealing algorithms, BP algorithms and niche technology are used in the method. The initial results of experiments indicate that this method can quickly find the satisfied network and that the network has good generalization performance. In general, GMNN can automatically design and optimize neural networks using the training sets.