Hi all,


I have researching neural networks, and also i implemented a backpropagation system, but the system where the ann acts its constantly renewed so retrain all the patterns for all new enviroments takes time ,and on a backpropagation neural network is delayed the process, i want to implement an ART neural network ,i should do it?

what's the question?

An ART implementation instead of Backpropagation

Ok maybe i did not explain my self very clear, i implement a Neural Network with a Backpropagation training algorithm but the application refresh or renew the information constantly,as backpropagation is an offline training algorithm and the patterns are very wide, the training takes a lot of time and the systems shuts down while is training the neural network, i understand that the ART Neural Network refresh its patterns online, the question is if the ART is an stable Neural Network and better for this cases?

If the ART NN combines learning with operation, then it might be worth looking into. Do you have any references for research on these?

ART uses unsupervised learning, which is different from backprop. You may want to consider SOMs (self organizing maps) if you're going the unsupervised route since there is better theory to describe what it's doing. Also, SOMs are fairly easy to implement. Basically they perform unsupervised clustering of your data.