本文为美国田纳西大学诺克斯维尔分校（作者：Philipp Koehn）的硕士论文，共67页。
神经网络和遗传算法显现出强大的问题解决能力。它们基于相当简单的原理，但利用了相关的数学性质：非线性迭代。反向传播学习的神经网络通过搜索各种函数来给出结果，然而，基本参数（网络拓扑、学习速率、初始权重）的选择往往已经决定了训练过程是否成功。这些参数的选择遵循实际使用经验法则，但其值至多是有争议的。遗传算法是基于选择、交叉和变异等原则的全局搜索方法。本文研究了遗传算法在神经网络拓扑优化等方面的应用，研究了各种编码策略对遗传神经网络协同效应的影响。根据它们在不同复杂度的学术和实践问题上的表现进行评价，并使用C++语言编写了一个研究工具，介绍了算法的基本性能。
Neural networks and genetic algorithms demonstrate powerful problem solvingability. They are based on quite simple principles, but take advantage of theirmathematical nature: non-linear iteration. Neural networks with backpropagationlearning showed results by searching for various kinds of functions. However,the choice of the basic parameter (network topology, learning rate, initialweights) often already determines the success of the training process. Theselection of these parameter follow in practical use rules of thumb, but theirvalue is at most arguable. Genetic algorithms are global search methods, thatare based on principles like selection, crossover and mutation. This thesisexamines how genetic algorithms can be used to optimize the network topologyetc. of neural networks. It investigates, how various encoding strategiesinfluence the GA/NN synergy. They are evaluated according to their performanceon academic and practical problems of different complexity. A research tool hasbeen implemented, using the programming language C++. Its basic properties aredescribed.
1 问题定义
1.1 神经网络
1.2 遗传算法
1.3 网络编码
2 多种方法分析
2.1 基本研究
2.2 直接编码
2.3 非直接编码
3 实验
3.1 问题描述
3.2 编码策略
3.3 加权优化实验
3.4 结构优化实验
3.5 结论
更多精彩文章请关注公众号：