RBF实现神经网络预测程序，压缩包，不用解压缩密码！

自组织系统Kohonen网络模型。对于Kohonen神经网络，竞争是这样进行的：对于“赢”的那个神经元c，在其周围Nc的区域内神经元在不同程度上得到兴奋，而在Nc以外的神经元都被抑制。网络的学习过程就是网络的连接权根据训练样本进行自适应、自组织的过程，经过一定次数的训练以后，网络能够把拓扑意义下相似的输入样本映射到相近的输出节点上。网络能够实现从输入到输出的非线性降维映射结构：它是受视网膜皮层的生物功能的启发而提出的。~..~-Kohonen network model. For Kohonen neural network, competition is this : For the "winner" of neurons c, in its switching around the region neurons in varying degrees, to be excited, and the switching outside the neurons were inhibited. Network learning is a process in the network connecting the right under the training samples for adaptive, self-organizing process, after a certain number of training, network topology can sense similar to the mapping of the input samples similar to the output nodes. Network can be achieved from input to output of nonlinear reduced-dimensional mapping structure : it is subject to retinal cortex of the biological function inspired by. ~ ~ ..

vb的波尔兹曼机（含模拟退火算法），有需要的可参考一下 -vb the Boltzmann machine (with simulated annealing algorithm), the need of a reference

线性神经网络应用实例-linear neural network application

因为递归程序运行的效率一般都比较低,本程序用java语言来实现TSP递归程序的优化-because recursive program run average efficiency are low, the procedure used java language to achieve TSP recursive process optimization

RBF网络用于函数逼近的一个程序！内部包含几个关于RBF网络的几个算法。-RBF network for function approximation of a process! Internal contains several of the RBF network of several algorithms.

运用神经网络的方法先将数据分类分析,然后再将未知类归类-The use of neural network analysis of data classification methods first, and then classify the unknown category

开发环境：Visual C++ .net2003 功能介绍：神经网络算法实验；分Console版本和MFC版本；主要用来求解TSP问题。-Development Environment: Visual C++. Net2003 function introduction: neural network algorithm experiments sub-Console version and MFC version mainly used to solve the TSP problem.

内附两个源代码，其中一个源代码是识别程序的，另一个是一个矩阵类库的，所以一并附上。 ~..~因为已经保存了训练好的网络权值，所以第一次使用识别程序时，可以直接运行Release目录下的可执行文件，然后对图片目录中的测试图片进行读入、识别。 当然读者也可以自行用训练样本训练网络，不过要特别注意训练样本的选择，否则可能造成识别率很低。训练样本选择的原则是，尽可能的有代表性，在训练时间不至于太长的情况下，训练样本数目尽量多。 -containing 2 source code, which is a source identification procedures, and the other is a matrix class library. So accompanied. ~ ~ .. Has been preserved because of the trained network weights, the first use of the identification procedures, can be run directly Release directory executable file, Then the catalog of tests conducted read into the picture, recognition. Certainly readers can use their own training samples training network, but special attention should be paid to the choice of training samples, Otherwise, it could cause a very low recognition rate. Training sample selection principle is that as far as possible be representative, in the training of the time not too long under the circumstances, the number of training samples as much as possible.

人工智能程序源码-artificial intelligence source

遗传优化神经网络结构，C++源代码-Genetic optimization of neural network architecture, C++ Source code