... With the OCR domain I'm using the solutions tend to not have any hidden neurons after pruning, which makes sense because I would expect it to be possible...
hey group, I'm having a terrible time searching the space for solving a cross product. Fitness is measured by giving it six inputs, one for x,y and z for each ...
If you are training on only a single example and every time you switch examples it goes back to square one it sounds like the problem is overspecialization....
Colin, Nice to hear you're getting good results. Would you mind reminding me what your representation is? 1) What is your training/evaluation set? ACSII?...
so in general, it's better to measure fitness in such a way that the fitness graph for the entire run is a single logarithmic curve, not repeated curves, with...
That is generally the way classification training works. I don't know any exceptions. The problem is if you only train for one example at a time, the system...
Hi Derek, ... I posted some details in an earlier message, but to recap I created my own clas (BitImage) to represent the characters during experiments as this...
... Ah...gotcha. Thanks for the recap. :) Yeah, I understand that this is just a sort of test domain. If you had any variance in your input, it wouldn't...
... The end product will use large images. But during the experiment, I use an expected output image size of 144x144, but the inputs are only 9x9 (plus ...
Things were going very slowly. The original seed topology (no hidden nodes) was still alive and kicking, and battling it out for first place in the ...
Hi John, I'm having a bit of a hard time wrapping my head around what you're doing. Is there any way you could post a couple of the compressed and target...
... Done. Photo album IEX has two images. Ignore the thumbnails, look at the actual images. The first image is a 9x9 input image, what is being input into...
... Well, it would require a more complex network to handle variance. Whether such a network could be found I don't know. ... Yeh, In formal NN lingo I think...
ken, right it's asymtotic, not log. Anyway, I set up a larger training set, but the fitness curve appeared to level off far short of close. It is still making...
... Yeh you need a spread of test cases that represent the range of input values you want the resulting network to accept, and you need to to test against all...
Chad, maybe you can post some examples from your training set? What are the inputs and outputs for example? It might give me some idea what's going on....
... you're ... compressed and ... look at the ... being input ... wants to ... pixel is ... 0,0 and Do you mean outputs rather than inputs? This description...
... hidden nodes) ... in the ... Last time I looked at your fitness graph it was going up; why do you feel things weren't improving? As long as things are...
Hi Group: My name is "Germán" it sounds like "her man" (but I'm not a playboy...unfortunately), I'm from Spain and I've been dedicated to AI as a hobby since...
... Yup...thanks for the images. Do you have any samples of enhancements by some of your better-performing ANNs that you could upload, to give us an idea of...
... No, I don't. I haven't written the code that does that yet. I've been putting it off because I can tell by the fitness values that it isn't able to do...
... Well, if there is, we'd like to find it. :) Floreano has published research in which a roving eye, or active vision system, scans a visual field in which...
... No, I mean inputs. What's being asked of the network is for it to form a model of its pixel inputs, and then use the top,left,bottom,right inputs to ...
hey everyone, i haven't been on in awhile but i just got started again working on my rabbit/wolf simulator. i was looking through christian's matlab code for...
... I created a function UniqueID(); I call it. It returns a number guaranteed to be unique within the run. It does this by just incrementing a counter and...
... I don't quite follow how this exponential increase in complexity comes about. I suppose it depends on how you set up the roving eye, I was thinking that...
Chad, Further to my previous email I was thinking about what sort and how many test cases you would need in order to have a good attempt at evolving a network...
Umm I'll try that picture again shall I :) y \ | / \ | / \|/ ... /|\ / | \ / | \ If it looks a mess then copy and paste into a text editor and change the ...
... John, Timmy, I suppose it depends on the implementation but keeping track of ALL innovations might become cumbersome during a long run. Perhaps then you ...
Having problems with message search?
Fill out this form to ensure your group is one of the first to be migrated to the new message search system.
