Well, I just received an assignment to create and train a neural network to predict the DJIA for a given day. The lecturer didn't have time to tell us much, so after about three hours of reading and poking about methinks what I want is a single-layer neural network (a perceptron network). It looks like something particularly suited to Prolog. What I don't understand is the implementation.

Here's what it *looks* like we need to do: (please, correct me if it looks like I've misunderstood)

• I take, for example, four parameters (we'll say the closing DJIA for the day previous to day in question, the average closing value for the week, average for the month and the consumer price index for that month, though this is somewhat arbitrary).
• To "train" the network (determine the weight of each parameter) I create a function which takes an arbitrary starting weight and refines it like so:
• W2 = W1 + (bias)*(last actual value - last predicted value)*(new input)
• Once the weights are determined (the network has been "trained") I have a new function to make predictions based on each of the parameters multiplied by their weights and summed.

Does that sound about right? What confuses me is how this perceptron is any sort of a 'network'.