Re: The A.I. Project
- Posted by dm31 at uow.edu.au Nov 07, 2002
- 497 views
Well. Anyways, I'll repeat what I said at first. We need a GOOD NN in EU first, if will will be using one in anything along these longs. NN's are more then just nodes with weighted inputs. That is the basic defination of a NN, but misses alot when I comes to making a useful NN Just doing this would be a big enough challange at the moment. We would need to think about such this as whether would would create a Feed-forward (associative) network or a Feedback (autoassociative) network. Then would we do single layer NN's or support the more complex multilayer systems. For the transfer function would we use either linear (or ramp), threshold or sigmoid?? Since we would most likey use a Adaptive network over a fixed network judging on what we are talking about would you use the Back-Propagation Algorithm (most widely used) for adjusting the weights as it learns? Will we have it setup for Supervised or unsupervised training? (yes CK, check it up, those are the two ways to classify an Adaptive network :>) These are questions and problems we most solve first. Lets not get to ahead of our selves at the start, it would be a shame if we started something like what ppl are proposing and it falls apart because we didn't think it through at the start :( I would love to see as build a project along some of the lines been suggested, lets just get the software we would need first. Cheers, Dan