... This sounds like an interesting task and a useful area to explore. It makes sense that as the "path" gets longer the ANN performs less and less well. I'm...
2649
Mitchell Timin
zenguyuno
Jul 1, 2006 4:46 pm
... You are cetainly correct that in the biological case, a one-bit error ruins the return path, since it no longer takes the animal home. And also, as you...
2650
rognorak
Jul 3, 2006 6:55 pm
Hello All, I stumbled onto NEAT a short time ago, and I think its an amazing piece of work - I'm not an expert on ANN's / Genetic Algorithms by any means, but...
2651
Colin Green
alienseedpod
Jul 3, 2006 11:17 pm
Hi Mitchell, All, I've intended to look into tasks requiring memory for some time now, and it should be really easy to throw together soem code for SharpNEAT...
2652
Derek James
blue5432
Jul 3, 2006 11:18 pm
... I'd recommend one of Ken's early publications where he first lays out the algorithm, like this one: ...
2653
Colin Green
alienseedpod
Jul 3, 2006 11:23 pm
Hi Mike, ... Currently there is no setting to do this so you would have to modify the code. One way would be to add the fitness threshold as a property on ...
2654
Mitchell Timin
zenguyuno
Jul 3, 2006 11:56 pm
... No, I started out with this one. But it's prolly a good idea to try the simpler one first. I'm not sure why I skipped that step. In my latest version I...
2655
Kenneth Stanley
kenstanley01
Jul 4, 2006 10:46 am
... by ... always ... might ... seem ... NEAT ... out the ... (Pubs)=114 ... To expand on to Derek's suggestion- although our papers on NEAT do tend to focus...
2656
rognorak
Jul 7, 2006 3:22 pm
First Off, thanks to everyone who replied to my questions, its good to see a newbie-friendly discussion group :) I have a lot of reading to do. In response to...
2657
Kenneth Stanley
kenstanley01
Jul 8, 2006 6:45 pm
Hey everyone, I wanted to post a quick message about what I've been up to recently. As some of you know, I started as an assistant professor at UCF 6 months...
2658
Kenneth Stanley
kenstanley01
Jul 8, 2006 7:11 pm
... good ... reading ... network, ... training? ... is ... to ... Mike, you can do that with NEAT alone. What you do is evaluate each generation on the...
2659
Daniel Larkin
daniel.larkin
Jul 12, 2006 4:33 pm
All, Just in case anyone might be interested, I have a paper on some of my work for a hardware accelerator for NEAT. It's available at the following url: ...
2660
John Arrowwood
jarrowwx
Jul 12, 2006 7:41 pm
All, For those that remember, I originally got interested in NEAT in order to do image enlargement. The primary thing that has stood in the way of progress, ...
2661
Kenneth Stanley
kenstanley01
Jul 13, 2006 2:19 am
I just found out that the version that was up of my tech report on CPPNs had been checked by a buggy spellchecker that let through a lot of typos. I never...
2662
petervermont
Jul 13, 2006 4:33 pm
I cannot see any files in the CVS folder on SourceForge for SharpNEAT. I know that the source is included in the package for SharpNEAT 1.04 but is there also...
2663
petervermont
Jul 13, 2006 4:57 pm
I have noticed that my best fitness sometimes declines. While I can sort of understand this if I have enabled 'pruning', it also occurs even when I disable...
2664
Colin Green
alienseedpod
Jul 13, 2006 10:02 pm
... Ok two points to start with: 1) SharpNEAT will always preserve the species with the best genome, even if it has reached the drop-off age. 2) Pruning isn't...
2665
Colin Green
alienseedpod
Jul 13, 2006 10:03 pm
Hi Peter, ... Sorry no. I may make the source available via CVS in the future, but right now it's an archive file only, plus there is a text file that ...
2666
Peter Andrews
petervermont
Jul 14, 2006 4:30 pm
Thank you for your detailed reply. I am very impressed with SharpNEAT. Thanks for sharing it. I do not understand how I am losing the most fit. My evaluation...
2667
petervermont
Jul 14, 2006 6:45 pm
I did some further testing and I can easily replicate SharpNEAT losing the most fit individual. To replicate: 1) Choose the Multiplication experiment 2) click...
2668
John Arrowwood
jarrowwx
Jul 14, 2006 9:36 pm
... That my not be a red herring. I'm not sure exactly how it might happen unless there was a bug in the code, but I can see where re-speciating an individual...
2669
Philip Tucker
tucker0171
Jul 15, 2006 5:38 pm
Hi Ken, I just finished reading your paper. Cool stuff. It's exciting to see you expand NEAT in such a novel way. I have a few comments and questions. 1) I...
2670
Colin Green
alienseedpod
Jul 15, 2006 9:25 pm
Hi Peter, Ok I was able to recreate the problem using the example you gave. John, yes it does seem to be a problem caused by the re-speciation and the example...
2671
Colin Green
alienseedpod
Jul 15, 2006 11:41 pm
... Ah yes, that old chestnut :) ... Seems like a completely reasonable and sensible approach to me. [...] ... I started out in EANNs by looking at incremental...
2672
petervermont
Jul 15, 2006 11:58 pm
Terrific! Thanks Colin. I am running with the new code now and will be able to confirm whether it fixes my original issue after running for a few hours. As I...
2673
petervermont
Jul 16, 2006 3:19 am
... I am using SharpNEAT and am new to all this so I would appreciate a little more detail. I have implemented two interface, IExperiment and...
2674
Kenneth Stanley
kenstanley01
Jul 16, 2006 4:03 am
Philip, thank you or your comments. It's really helpful to hear thoughts like yours at this early stage of research. I'll try to address some of your...
2675
Kenneth Stanley
kenstanley01
Jul 17, 2006 3:35 pm
Hi John, I'll try to answer a couple of your questions below. By the way, I had a student in my class who used NEAT for image enlargement (he called it "image...
2676
Daniel Larkin
daniel.larkin
Jul 17, 2006 3:46 pm
All, A short question... It seems to be accepted that regularisation methods (weight decay etc) can be beneficial from a generalisation perspective in ANNs....
2677
John Arrowwood
jarrowwx
Jul 18, 2006 3:50 am
... Correct. Based on what I have read, an appropriate subset is 9x9 pixels at a time. As a general question, in what way will NEAT inspired interpolation ......
