... dealing with AI. ... fundamentally conceptually flawed, and that these flaws often stem from an oversimplification of the meaning of "intelligence" and/or...
... Yeah, I thought the same thing. Of course, it is easier to criticize a given approach then to put forth a comprehensive one of your own. And much of the ...
I'm interested in exploring the limits of what ANNs can do, so my next problem is going to be one that's well suited for the normal computing paradigm, but not...
... Hi Mitchell, concentration does sound like a very interesting problem for an ANN. It is probably quite difficult since it involves long term accurate and...
... OK, done. There are two ASCII files, concen.c & consubs.c. They are to be compiled and linked with any C compiler. Yes, I think this will be very ...
... Inquiring ... Hi Mattias, how are you doing? I am engaded in three main projects right now: 1) We are developing a video game based on neuroevolution....
... I agree with Ken that the representation of the cards is a major issue, but I'm afraid I don't quite understand your explanation of how you're handling it....
... And you've done experiments both ways, correct? Did you find one approach to be more effective than the other, or did it depend on the domain? Derek ...
... <snip> ... We don't have to represent suits because they have no role in concentration. I'm planning to start with a very small deck and work up if I'm...
... Well, they could, if you are using a standard deck and, for example, only red or black pairs could be matched (there are lots of different variations that...
... If you change the rules to require color matching, it just means that there are twice as many kinds of cards, but half as many of each kind. It still fits...
... What I was describing wouldn't build in any particular strategy. It would simply mean that the input to the ANN would be two board positions each turn,...
... <snip> ... I guess I didn't understand your proposal; I'm not sure I do now. Are you suggesting that we only inform the ANN of changes to the tablaeu of...
... Yes, something like that. I'm suggesting that the inputs to the network each turn are only for the two positions on the tableau that it is "choosing". For...
... cards ... not ... indicates ... I have the same feeling, since if you combine nodes then somewhere in the net it has to separate out the information again,...
... Hey Derek, is the idea here that the X and Y coordinates would be "picked" or just looked at next? In other words, is the net "looking around" at all the...
... <snip> ... <snip> I read through that thread. Now there are three ways of doing this! :) Originally, I had only considered looking at the whole table,...
... I was thinking it would be simpler to just have the X and Y coordinate output be the next cards picked. I'm not sure in Concentration that there needs to...
... Right...these are basically very similar approaches, differing primarily in the amount of information input into the net each step. ... Yeah, in this case,...
... <snip> ... That would correspond to an illegal or invalid move. My approach to evolution is very draconian; if a game playing ANN wants to make an invalid...
... I'm also not sure you'd want your hidden units to be binary. Making them binary just means that for certain functions, which we don't know whether or not...
... But learning what is legal and what it illegal can put an extra strain on the already difficult problem. For example, a network that is doing quite well...
I am interested in how NN can be used with GA,GP in evolve real-time decision making bots. I look forward to reading the messages from this group and getting...
... I agree with Ken that a "one strike and you're out" selection pressure against illegal moves probably isn't going to be very fruitful. I like his approach...
My preference for the Draconian approach is based on limited experience. I had good results with it for my Nim evolver. There I was using floating point...
... Hi Rog, have you seen the demos at: http://www.cs.utexas.edu/users/kstanley/demos.html They show simulated bots that word evolved for real-time decision...
... group ... Hi Roger...definitely check out Ken's bot experiments. He's done some interesting stuff with real-time decision-making bots. I'm currently...
