Hi guys, Any help greatly appreciated, I'm having real difficulty trying to work out how neural networks work. So if you are given a training set of data where the class is known, you: Sum up all attribute values for each item in training set Squash the value So this number that is given after this process? I don't get what it's for. So if you are given some data with an unknown class, you pass it through the neural net and do you see which value it's nearest and take that class?

ok...

a node have 5 other nodes as input.
each of the 5 connections have a weight.
you multiply the input nodes values with the connection weights, and then
pass this new number trough a "neuron" function; for example sigmoid.

The interesting stuff happens when you connect them together into a net.

Read: (a suggestion on learning order)

#1 (1 hour)
Example of AND/OR/NOT/XOR (this will show you that you can represent logical rules with the ann)

#2 (1 hour)
ANN Backpropagation (this will show you how to train a network)

#3 (5 hours)
Make a small ANN that learns the logical operators (AND/OR/NOT/XOR) with backpropagation

#3 (1 lifetime)
Experiment with different problems and different implmentations to build yourselfe a better understanding of what you can use ANNs for.

Pros:
ANNs can handle noize (successfull experiments exists)
ANNs can handle missing data (successfull experiments exists)
ANNs can find patterns in what appears to be noizy data (successfull experiments exists)
ANNs can transfer its training data to new and unseen problems (successfull experiments exists)
ANNs are fast when trained (successfull experiments exists)