>he doesn't program evolving systems It's the most

I'm working on a cellular automaton because I'm putting together a portfolio to support a masters application. I'm also going to model bird flocking, ant signalling and something to do with genetics since I want to do bioinformatics. If I have time maybe I'll do something like this.

>>58770307

>cellular automata able to evolve its rules, to become as non-noisy as possible
>evolve its rules

thats very interesting, are there any papers on those kinds of systems

>So that it can learn to learn
That's a human anon, if you manage to do it you are the richest mofo in the world

>>58770307
Maybe you can program a neural network to try and cure your autism

>>58770390
If you go 1 step further you have God. Learn to learn to learn

>>58770307

>to become as non-noisy as possible

how did you quantify noise in the simulation?

>>58770390
I can't imagine it's too hard in theory.
You just need an appropriate protocol for neurons to create more neurons and link with each other.
In practice, it'd take loads of spamming randomly generated neural systems before you got something that you could start to evolve.

>>58770556
For every line, both vertical and horizontal, count the number of sublines (a line of cells the same colour).
The fewer sublines there are, the lessy noisy it is.
The ratio of the two colours was also a factor, to prevent it from all going one colour.

The ultimate form with those rules is to have a white rectangle and a black rectangle of equal size.

>>58770685

can you post some screenshots demonstrating this

>I can't imagine it's too hard in theory.
>just need an appropriate protocol for neurons to create more neurons and link with each other

decades of research has gone into this and its not that simple, "learning" means an ability to infer, store, and interpret many kinds of information vital to survival and or reproduction, the brain works the way it does not because of neurons alone, but because of millions of years of epigenetic structures and the information passed on through culture, the way you've trivialized this communicates you really need to learn more about human cognitive processes as well as some of the problems in AI, like general learning

>>58770796
I don't imagine I'm going to make a revolutionary change overnight, but I do believe that the training program I want to try out is very interesting.

The training program I have in mind is to pit two AIs against each other, with each of them having two screens they can use a subset of neurons to control.
One is a colour screen, and one is a betting screen.
The betting screen allows the AI to bet on each pixel of the colour screen, and what it's betting is on how closely that pixel will match the same pixel of the opponent's colour screen.
There's all sorts of variants of the betting system that could be tried out, for example whether it's betting a given amount with the reward/penalty dependent on the similarity of the pixels, or whether it's betting on what the difference will be with the reward dependent on the accuracy.
The AI gains score by guessing well, and there is an input neuron to tell it how much score it's gaining to save it from having to understand the rules.
There's also input neurons so it can read the opponent's colour screen, maybe their betting screen too.

Obviously, this will result in the AIs simply mindlessly translating the opponent's colour screen into appropriate bets, since the betting isn't turn based and so guessing accuracy determines the rate of score gain.
Unless, of course, the AI is slowed down so it is forced to predict!
If the neurons were given points in space, then perhaps the distance between them could determine how long it takes for a signal to travel.
The inputs and outputs would be fixed in space, so the AI can't just squish everything into a tiny cluster and not need to predict by operating extremely fast.
That could have an interesting effect: It's possible that the AI's neurons will then form multiple tight clusters, which carry out different processing tasks.

>>58771122

again, I'd like to reiterate that, while its good that you've taken steps to play with programming these systems, it would do you some good to read the literature on AI as well

>That could have an interesting effect: It's possible that the AI's neurons will then form multiple tight clusters, which carry out different processing tasks.

well you're programming a computer, the output of your program, even if "stochastic" elements are involved such as random number generators for your simulations or whatever, is deterministic on the formal structures emergent from your programmed systems: a lot of literature has already gone into studying this and you may be repeating, poorly, what someone else has formalized well

toying with math is fun, but if you actually want to learn how to do thing wells you need to be rigorous and learn from others

>>58770307
>why would you not?
Why would I waste my time doing something useless? Coding is my job - I would rather code something I'm getting paid for.

>>58771504
You could run a genetic program to get the job done for you if you have enough time to let it run and you have a non discrete problem space. Oftentimes that's going to be superior than anything a human can cook up anyway, e.g. Traveling Salesman.

>>58772438
I don't think a generic program would set up a server, debug a project, fix design of an app and configure a bunch of VMs for different purposes.

>>58772498
>and you have a non discrete problem space

>>58772677
I suggest you do your homework.