For neural networks learning a Backpropagation algorithm works best see http://en.wikipedia.org/wiki/Backpropagation for a good introduction.

Well to answer this question you have to understand what :

1) A simple perceptron does (it finds a linear separation for your classification problem)

2) A multiple layer perceptron does (it uses the hidden layer as a transformed space to turn a non linearly separable space into a linear one)

3) What the specific learning method (in your case probably backprop) does. Some learning method will work best for some problems. An important factor is also the way you code your input (i.e will you encode the value "3" in a single cell in [0,1], in 3 cells (0 0 1) or any other form of coding. This coding will affect a lot the type of problem your perceptron can learn or not.

If you want to learn how an ANN work and how someone can code it, I think the best place is Matlab help. Check the ANN toolbox in Matlab, read the pages. It tells you enough information to understand ANN and there are many examples on that.

A bit of literature might help - see attachment.

Hi,

this is relevant if you are looking to actually understand the algorithms used to train neural networks.

I like the UFLDL Tutorials (http://ufldl.stanford.edu/wiki/index.php/UFLDL_Tutorial). They focus on autoencoder networks, but they teach you how to learn any multi-layer network and you get Matlab codes to get you started.

Geoffrey Hinton has a course on neural networks on Coursera https://www.coursera.org/course/neuralnets, perhaps you can access those videos.

Andrew Ng has machine learning classes on Coursera wich include neural networks https://www.coursera.org/course/ml.

An interesting tool is Theano: http://deeplearning.net/software/theano/tutorial/index.html#tutorial

Matlab NN toolbox manual is a good option.

Matlab Neural Network Toolbox

Dear Rodrigo,

I agree with the previous comments that Matlab has an incredible help where you can grasp most of the ideas behind NN multilayer perceptron and it will suffice for a first approach to this field. When you really want to get into the NN world (beware, it will cost you a certain effort) I would suggest one of the most appreciated book in this field:

Neural Networks: A Comprehensive Foundation ( http://www.amazon.co.uk/Neural-Networks-Comprehensive-Foundation-International/dp/0139083855).

P.S.

Probably there is a way to read, at least part of it, online for free.

The following website might help, where you can get some references to paper on developing ANNs and download an Excel add-in implementing ANNs: http://www.ecms.adelaide.edu.au/civeng/research/water/software/neural-network-excel-add-in/

For theory Part I would suggest to read ANN Yegnanarayana or Wiki or courses Like MIT OCW would help.

To experience it Matlab toolbox is the best .