Good point. I think it will test the ability of the NN to distinguish between different types of features. In the example I gave, the fact that there are three different shapes, not two, or that the shapes are not divided into three of one kind and one of the other, is pertinent information, if the NN is capable of recognizing it. Bear in mind that the NN I envision is capable of doing completely different tasks. I might ask it to repeat some pulses, and then immediately after ask it to do an odd-one-out problem. Perhaps the NN will learn to distinguish between different types of odd-one-outs: it will recognize whether it is being asked to do a colour comparison, a shape comparison, a size comparison, etc. This seems fairly plausible for even quite a simple NN to accomplish, and I would be disappointed if mine couldn't. This would be correlated to several pathways being stimulated in parallel - one for each task. The winner - the one that generates output - would be the one that is most stimulated, which would correspond to a good match. What I would really like to see is a NN solving the general odd-one-out problem - so that the first time it sees one based on size, say, it is more likely than chance to be right, and remembers even from that one example if given positive reinforcement. This would probably be correlated to some kind of novelty detection, interacting with pre-existing groups of neurons customised for size detection. I want to clarify, as well, that the 'evolution', while always present to some degree, is intended mostly to shape the network into a highly adaptive configuration. I would expect it to come into play if I added a new modality like vision, but not to have as great an effect when a new problem is encountered. I would expect the genes already present, and the pathways already present, to be well equipped for learning such a task.
On 12/27/06, Mike Dougherty <[EMAIL PROTECTED]> wrote:
in general, how do we indicate the odd one out of that set? Sure it's "obvious" that the color is important in this case - but I see two circles and that the square is more similar to the circle(s) because of the higher number of sides. Therefor the triangle is the "odd one." What rules does an evolving neural net use for determining the pattern in order to determine the exception to the pattern? On 12/26/06, Nathan Cook <[EMAIL PROTECTED]> wrote: > > The training set should have problems of (at least) two forms to test my > hypotheses: > (1) after 'hearing' a sequence of pulses, reproduce them, and (2) after > being presented with several images (e.g. red circle, red square, red > triangle, green circle), indicate the odd one out. Being able to do either > of them should show I'm on to something. > Is that any help? I was reluctant to give too much away because it's a > rather far fetched concept, but as you can see, the neurons have to be > capable of doing a lot. I think I can justify taking this one of many > options in neural networks, if only because no-one seems to have let the > neurons themselves compete before. > ------------------------------
----- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?list_id=303
