Hi all, Well my project is not finished yet, but here is something that I want to share with the community. It is about using CPPNs to grow physical or...
Seeing Peter Krcah's stuff (http://ero.matfyz.cz/) at GECCO this year got me re-interested in artificial embryogeny. I went back and reread Ken's paper "A...
Hi, I'm not sure if it is the answer to your question... A classical benchmark is pattern formation, where the pattern is a 2d picture. - J. F. Miller,...
Christian, Correct, the semaphore is just for decoding the genomes produced by the substrate, because genome decoding is not thread safe. Since you're not...
Hi Peter, I've also started playing around with CPPNs whose output change over time as well. This sounds like a very interesting idea and I'll be interested...
Hi Derek, I'm not really aware of much, if any benchmarking work that has been done in the field other than some of the papers mentioned in the other post. An...
Peter, I'd be interested to see some of the objects that you grew. By the way, another way to create contiguous objects might be to evolve them as connectivty...
Derek, I also don't believe there has been a great deal of benchmark comparison sine the 2003 paper, although Alexandre Devert cites a few. However, usually...
... Right. It seems like there's a lack of rigor in determining the representational power of these systems, and this is what I liked about the way Ken...
... But isn't that the general case of getting stuck in a particular part of the solution space? If a particular candidate repeatedly fails to represent very...
Hello, I want to discuss with you about this topic. The main bottleneck of NEAT is the evaluation of every created network to get the fitness. But what if you...
Hi Christian, ... You could take the Pearson correlation of the data sets. The algorithm is fairly simple for calculating correlation for two variables, x and...
Hi Christian, I did some experiments with similar stuff a while ago. Some of them were related to Neat, others were "plain" GA, but the situation was the same:...
Hello Sandor, thank you very much for your detailed post. Testing the younger generations with less training data sounds very good. But currently I am using...
Another possible option would be to use a technique called Spherical Exclusion to partition your data. The premise is that you chose a data point at random...
Thank you Robert, that sounds exactly like what i am looking for! I also understand the theory, but have problems with the mathematics behind you theory. So I...
I would suggest ignoring the output at this point and looking only at the input variables. Here's a quick bit of pseudocode that may help: 1) Pick a single...
I forgot to mention that during the calculations of distance, you could omit taking the square root. If you have a very large dataset, the square root ...
Thank you very much, your pseudo code and formula helped me a lot (Now I understand the mathematics) I will try it. It looks very promising! It will speed up...
... Just FYI, IGenome.GenomeAge will give you the age which you can do in the population evaluator and pass to the network evaluator with an overload of...
Hi all, I'd like to compare NEAT with my (of course new and powerful) neuro-evolution methods. I've worked with the Boolean function (4 inputs, 1 output) : (A...
Ken gives some tips about tuning your system to work with XOR here: http://www.cs.ucf.edu/~kstanley/neat.html under "How should I test my own version of NEAT...
Jean, give us the input/output table and the metric used to compute the fitness. I can think of many different input sets for your example, I just need to make...
Hi David, I leave the cells to grow to any X,Y. A cell is basically one fixed- size square in a grid. Cell division tries to put a new cell directly next to it...
Hi all Sorry its taken so long to reply - day job keeps me v busy (oh and the kids too!). I wrote a brief paper for the competition (a Word doc) that explains ...
Hello everyone, I am about to head over to the University of Sussex (in Brighton, England) to do their MSc in Evolutionary & Adaptive Systems. However, as I am...
Mike, I'd guess there are others in the group with similar questions so I think it's probably useful to the group. (By the way, by "Adaptive Networks," I...
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.
