No replies to this topic

[rouncer]

Why havent i coded my first perceptron yet?

Because i cant get over the fact that they are learning disabled, they learn through trial and error and you need a million going at the same time to test for the most successfully linked net, and that has practicality problems... that i want to fix, and would bring it closer to how we operate.
Even though theres been fantastic results with what they have already, and the practicality problem is removed when operating in the virtual realm, still it annoys me that it could be better.

What foggy ideas do you have...
Well, i think that it needs some way to "build" on what it last had, instead of completely wiping it each new instance.
also, teaching it language has a moderation problem, its hard to write a moderator program for it, youd have to sit there learning in slow mo telling it if it said real english or gibberish, but if it had this "building" mechanism, then you could hand teach it!
and it could learn to walk without over trialing.

So... they over trial, and you dont like it.
I wonder if im onto something...
That we can reduce trials, is it still possible to reduce trialing?

If it didnt overtrial, it could be feesable to teach it language.