The Game of Go has been explored for 2500 years and resulted in a body of theories that Go masters study in Go schools. Alpha Go discovered that human Go understanding was caught in a local optimum, and broke out of it. The game of modern physics started only 130 years ago...
-
-
Replying to @Plinz @gegarciaar
physics is not a game with rules, it is about discovering the rules of the game.
2 replies 1 retweet 28 likes -
Replying to @SergioFelperin @gegarciaar
Consider the Atari player that preceeded Alpha Go: it was not given the rules of Donkey Kong either. We are always approximating a function to optimize some loss criterion; it does not matter if the loss represents game score, prediction accuracy or model size etc
2 replies 0 retweets 2 likes -
Replying to @Plinz @gegarciaar
in more metaphorical terms: can an algorithm find alexander's solution to the gordian knot problem?
2 replies 0 retweets 1 like
It comes down to discovering that your current description of the problem space constrains you to a bad local optimum in solution space, so you optimize instead on the level of the function that generated the problem space abstraction (= find different model over known patterns)
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.