Many people have tried to defend pure DRL w Nature Machine Intelligence article that actually is a hybrid model; the below youtube pointer IS to a pure convnet - but read fine print: “network works decently well for any position less than 6 moves away from solved”
#symbolphobiahttps://twitter.com/AlexRoseGames/status/1186571935611850752 …
-
To solve Rubik's cube in generality, you need to approach it as a reinforcement learning problem. You must explore the search space somehow, e.g., MCTS, Thompson sampling, etc. Is all search in discrete spaces "symbolic"? Not in an interesting way
what do you mean here by “interesting way”? eg MCTS requires a full set of symbolic resources (symbols, variables, instantiation, operations over variables). how can asymbolic system reliably traverse trees? there may be some asymbolic way of performing search; MCTS isn’t it
-
-
MCTS can treat states as atoms as long as it has a successor function that maps atomic states (plus actions) to atomic states. So it doesn't make use of any of the compositional power of symbols, which is what I meant by "interesting"
-
well, i asked about operations over variables, which is what i think is even more foundational. thoughts there?
End of conversation
New conversation -
-
-
I personally agree that MCTS is a symbolic planning algorithm, even when you plan in domains that have a somewhat opaque forward model (e.g. ALE or GVGAI). But I suppose most people think of symbolic methods as involving a language with predicates, e.g. first-order logic.
-
i am most concerned with quantified statements, which seem to me to demand “rules” or operations over variables. can you build MCTS without them? (nobody does in practice AFAIK)
- 4 more replies
New conversation -
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.