Hi,

We have to select a project for our 3rd year engineering, and we have decided to implement a system which recognizes the language (from a sentence, paragraph, or a complete long text). We aim to do that using neural networks.

I am absolutely new to 'world' of neural networks. Is the project I have described above feasible? I have absolutely no idea of neural networks currently. We intend to 'train' the network using passages of languages first (English, Spanish, French etc), and then make it ready for recognition - is this how things work? Can someone please guide us further how to proceed?

We can use any software to implement this - MATLAB, C/C++, Java.. What would be the best tool for a newbie?

Thanks.

PS: Pardon me if this post is in the wrong forum, but I could not find any forum on PH dedicated to neural networks.

I think that matlab is the best tool for a newbie because it is an high level lenguage. In matlab, for example, if you want to create a neural network there is a matlab function that can create it. Morover there is a well done help with a lot of example.

Thanks giug :)

Can you please give me further details - how do I start out with NN? What theory do I need to know before I can start implementation in MATLAB? Are the tutorials of MATLAB enough for a person who has no idea of NN? Do I need to know about semantics and language processing for this project?

Thanks.

On internet there are many sources about neural networks.
You can begin reading the matlab help, write in the help search "neural network getting started", there is an introduction about the neural networks and how you have to use matlab to implement neural networks. Then, obviously, you have to know some matlab sintax, but, for this, in the matlab help there is all you need.

Thanks a lot once again, I'll start out with the MATLAB help files :)

MATLAB is convenient for programming neural networks and other machine learning algorithms, especially because it allows vectorized code, like this forward pass through an MLP (the most common type of neural network):

% forward pass in MATLAB
HIDDEN = tanh(HIDDENWEIGHTS * INPUT);
OUTPUT = tanh(OUTPUTWEIGHTS * HIDDEN);

This code calculates the neural network output for all observations, not just one. Goodbye, do loops!

One place you could start is my DeltaRule routine, which trains a single artificial neuron using the delta rule:

DeltaRule posting


-Will Dwinnell
Data Mining in MATLAB