Machine learning method generation plausibility.

[Etotheipi]

I don't know very much about machine learning, but do any of you think it would be possible to generate human usable methods with a machine learning ai?

 

[Hazel]

I think it would be difficult (but not necessarily impossible) for a machine to be able to tell the difference between a method that is good for human solving and one that isn't... interesting idea though!

 

[Etotheipi]

I think it would be difficult (but not necessarily impossible) for a machine to be able to tell the difference between a method that is good for human solving and one that isn't... interesting idea though!

i suppose a program that gives a score.to the ergonomics would help, and maybe try and make the program slant towards block based methods so it doesnt have like 10 pieces solved in random places and then filling in the rest.

 

[GAN 356 X]

i suppose a program that gives a score.to the ergonomics would help, and maybe try and make the program slant towards block based methods so it doesnt have like 10 pieces solved in random places and then filling in the rest.

That would be pretty cool. Putting CFOP, Roux, and ZZ through it would be interesting

 

[Etotheipi]

That would be pretty cool. Putting CFOP, Roux, and ZZ through it would be interesting

Yeah, maybe ill try and do something with the extremely tiny bit of python knowledge i have. =D

 

[TipsterTrickster]

someone did this with 2x2, but it could probably be explored waayyy more than this.

 

[Underwatercuber]

Biggest factors I can think of that it would have to consider is recognition, ergonomics, move count and alg count. If it can manage those then it can probably make a decent method.

 

[PapaSmurf]

I don't see why it wouldn't be plausible at all. They've already made a machine that has learnt to solve a 3x3, so it'll be expanding on that idea to find methods that work for humans and are actually good.

 

[Etotheipi]

It would also be interesting, but probably quite unlikely, if a program generated a method similar to a human made speedsolving method.

 

[Deleted member 36445]

I don't see why it wouldn't be plausible at all. They've already made a machine that has learnt to solve a 3x3, so it'll be expanding on that idea to find methods that work for humans and are actually good.

Solving the Rubik’s cube with deep reinforcement learning and search - Nature Machine Intelligence

For some combinatorial puzzles, solutions can be verified to be optimal, for others, the state space is too large to be certain that a solution is optimal. A new deep learning based search heuristic performs well on the iconic Rubik’s cube and can also generalize to puzzles in which optimal...

www.nature.com

 

[ImmolatedMarmoset]

There was a thread about this a few months ago that kinda died.. lemme see if I can find it