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 …
-
And most search algorithms, if not all, use comparisons to search a state space. I bet comparisons operate on symbols. So to do search there must be some form of symbolic representation. Right?
one could argue that deep learning does a kind of search; standard search systems certainly are sets of symbolic operations over variables
-
-
Yes, but the search space is over the parameters of the network. I don't know anyone who has managed to embed state space search (a la Rubik's cube) into gradient search in weight space. Maybe someone has tried to create differentiable approximations of such spaces?
-
A few people have tried to learn programs from data in this way. They embed the search problem over programs as gradient search in weight space. If you haven't seen it, you might find this paper interesting. I found it a useful review: https://arxiv.org/abs/1611.01988
End of conversation
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.