In short - more optimisations, some important bug fixes, and a new 'Update Frequency' menu on the main GUI. Zip file and more details in the usual place: ...
Hi: I'm currently working on a modification to NEAT (using a genetic regulatory network to create a more flexible genome), and I had a question about parameter...
... That's the $64,000 question, isn't it? :) You say you'll give bonus stars for non-ad-hoc approaches, but as far as I know, that's the way most people tune...
Hi Bryan, ... This sounds very interesting, do you have any more details worked out yet? Also, I was wondering (since you're at MIT) if you were working with...
Bryan, Derek and Joe make some good points. I'd like to add one other thing about parameter tweaking which is that sometimes you can look at the bigger...
The discussion about parameter settings coincided with, and has some bearing on some thinking I've been doing about connection weight mutation schemes. This...
Hi, Colin, It's an interesting idea, that you proposed, but I would like to note that there is one more way to try to tune some GA parameters (weight mutation...
... Yeah, our mutation rates are basically the rate at which new topological mutations arise among all possible locations where a legal mutation could occur...
Hi Colin, Another take on this is something I've pondered about for GP, and which is inspired by nature. In nature, not all parts of a chromosome exhibit...
... Hi Yuri, I did consider this approach, one problem I can see is that I think lot more runs would be needed to seperate out genuine positive results from ...
... Given these facts it seems highly likely that my initial random population of neat parameters will all perform about the same and that there will be no...
... See my reply to Yuri! ... I think we're all getting carried away a bit here. As Derek points out, the parameters probably don't effect a search that much...
... I was just scanning over comp.ai.neural-nets and came across FANN, the Fast Artificial Neural Network Library. It looks like a good quality piece of work...
Colin, one problem with your meta-search for the right mutation parameters is that it is likely if you find anything at all it will be domain-specific. The...
... Not only that, in a larger sense, different phenotypic modules exhibit different mutation rates based on their connectivity. For example, take some...
On Mon, 11 Oct 2004 19:36:33 -0400, Joseph Reisinger ... Um. So perhaps mutation rate could be applied at a level below the whole individual but above each...
... I think this is different (and maybe thus different from what you want to look at right now). There's two related questions: 1) f-- Sorry, there's three...
As a side note, what I hear in all this discussion is an effort to sort of re-invent backpropagation... The point being that if we take a moment to recognize...
... Not only domain-specific, but it is inside each search, rather than an externally determined "ideal" value. E.g. as I mentioned in my previous mail... ......
... Interesting point. I wonder would there be a benefit to be had (in some problem domains at least) from applying a modicum of backprop to a NEAT network?...
... Hi Ian, Sorry, I meant in biology this kind of thing is common. I think its important to point out this difference though, since a direct-development...
... I do work with Una-May, in fact, she's sort of my de facto co-advisor (Rod Brooks is my nominal advisor). She's very helpful, especially since I'm the...
... Sure...let me quote from one of your posts... ... Basically, I was saying what you were saying...that this takes us one step closer to back-propagation /...
... I've not talked to her but I saw some of her work at GECCO this past June. Also, I'm looking to apply to graduate schools this winter and the lab at MIT...
... I agree, and if this is the case we can then look at the numbers that come out of each domain and build up a sort of lookup table that we can refer to...
... Ok to clarify, Yuri suggested meta-params on each genome, and you are talking about an even finer level of granularity with meta-parameters on each gene or...
A better (and simpler definition) of ES: http://computing-dictionary.thefreedictionary.com/evolution%20strategy Whcih I may as well quote in full: -- cut here...
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.
