径向基网络字符识别 [英] Radial basis network character recognition
问题描述
我想通过实现一个给定的神经网络样开发一个简单的字符识别程序;一个简单的命令行式就足够了。
I want to develop a simple character recognition program by implementing a given neural network kind; a simple command line-type is enough.
径向基函数神经网络被分配给我，我已经研究了重量训练，投入到隐藏到输出的过程，但我仍然怀疑，在执行它。我引用（1）和的（2）。
The radial basis function neural network was assigned to me and I already studied the weight training, input-to-hidden-to-output procedures but I am still doubtful of in implementing it. My references are (1) and (2).
10×10的二进制对象（即重新presents一个字符）的简单的一维数组为输入。例如，下面的阵列
A simple one-dimensional array of a 10 by 10 binary object (that represents a character) is the input. For example, the array below
input = array(
0,0,0,1,1,1,1,0,0,0,
0,0,1,0,0,0,0,1,0,0,
0,1,0,0,0,0,0,0,1,0,
1,0,0,0,0,0,0,0,0,1,
1,1,1,1,1,1,1,1,1,1,
1,0,0,0,0,0,0,0,0,1,
1,0,0,0,0,0,0,0,0,1,
1,0,0,0,0,0,0,0,0,1,
1,0,0,0,0,0,0,0,0,1,
1,0,0,0,0,0,0,0,0,1 )
是字符A的再presentation：
is the representation of the character "A":
0 0 0 1 1 1 1 0 0 0
0 0 1 0 0 0 0 1 0 0
0 1 0 0 0 0 0 0 1 0
1 0 0 0 0 0 0 0 0 1
1 1 1 1 1 1 1 1 1 1
1 0 0 0 0 0 0 0 0 1
1 0 0 0 0 0 0 0 0 1
1 0 0 0 0 0 0 0 0 1
1 0 0 0 0 0 0 0 0 1
1 0 0 0 0 0 0 0 0 1
我打算采取输入的总重量，并比较它的训练设置为在字母表中的其它字符和一个与最接近所保存的1-D阵列是prediction
I plan to take the total weight of the input and compare it to the training set as in the saved 1-D arrays of the other characters of the alphabet and the one with the closest is the prediction.
现在的问题是我倾向于理解算法更好，如果psented在CLRS-方式或类似$ P $，而不是数学公式。我很难理解在这两篇文章的解释（我觉得这是最简单等等看这里的谷歌搜索）。
The problem is I tend to understand algorithms better if presented in a CLRS-manner or similar type as opposed to mathematical formula. I find it hard to understand the explanations in those two papers (which I find the easiest to read among others here in the Google search).
有人点我到一个友好的算法，一个RBNFF这需要在一个数组中，并产生权重的输出？如果不是这样，一份文件，解释了这种外行的方式将AP preciated。
Can someone point me to a friendly algorithm for a RBNFF that takes in an array and produces an output of weights? If not, a paper that explains this in Layman's manner would be appreciated.
推荐答案
有关我能找到有没有一个正确的方式来训练他们。
Training
For what I could find there is no "one right way" to train them.
最简单的训练我能找到的是由两个算法组成
The simplest training I could find was by a composition of two algorithms
-
（聚类）以在网络的左侧部分（输入权重和的RBFs）和做无监督聚类。有几件事情，你可以尝试硬/软和集群/的RBFs数量。 每个群集是权重连接到一个单一的RBF的重新presentation。 如何您不必集群得到RBF和RBF的权重去取决于什么集群所使用。 （我可以延长这个，如果还不清楚）
(Clustering) Taking the left part (input weights and RBFs) of the network and doing unsupervised clustering. There is a few things you can try out hard/soft and the number of clusters/RBFs. Each cluster is a representation of a single RBF with the weights connecting to it. How you go from having clusters to get rbf and rbf weights depends on what clustering you are using. (I can extend this if it's unclear)
（神经网络）的利用从聚类的输出作为输入到一个普通的单层神经网络求解从最后步骤的原始RBF网络的遗漏部分
(Neural Network) The solving the left out part of the original RBFNN from the last step by using the output from the clustering as input to an ordinary single layer neural network.
可能更容易找到这些更原始的算法容易解释
Probably easier to find these more primitive algorithms easily explained
发现了一些伪 - code与解释这或许可以解释这一切更好的（C＃编写的）
found some "pseudo"-code with explanations that might explain it all better (written in C#)
http://msdn.microsoft.com/en-us/magazine/ dn532201.aspx
（据说）工作蟒蛇code
(Supposedly) working python code
https://github.com/andrewdyates/Radial-Basis-Function-神经网络
这篇关于径向基网络字符识别的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
