satisfactory explanation to why some moves feel good - they just do, based on pattern recognition built up from countless previous games.
-
-
Replying to @xuenay @Intrinsic29
I don’t think “we don’t know in detail how it works” is either interesting or meaningfully similar to humans (though also true of us). YMMV!
1 reply 0 retweets 0 likes -
Replying to @Meaningness @Intrinsic29
Most relevantly it was a capability AI didn't really have before; we can train humans by throwing them lots of training examples and without
1 reply 0 retweets 0 likes -
fully understanding *how* they learn; now we can do same with AI. That opens up huge practical applications, we need to understand less.
1 reply 0 retweets 0 likes -
Replying to @xuenay @Intrinsic29
This isn’t new at all. Machine learning has been doing this since the 1950s.
2 replies 0 retweets 0 likes -
Replying to @Meaningness @Intrinsic29
1950s ML didn't replace the other, more knowledge-intensive approaches the way DL is replacing them now, nor be capable of doing as much.
1 reply 0 retweets 0 likes -
Replying to @xuenay @Intrinsic29
The DL algorithms are little changed from mid-80s backprop which we ran on 1MHz processors. The main difference is GPU power. Brute force.
2 replies 0 retweets 0 likes -
Were that true there would be no difference in using different architectures. Everything would just be larger MLPs.
1 reply 0 retweets 0 likes -
Yes, there has been an accumulation of “art, not science” knowledge of how to tweak architectures and hyperparameters.
1 reply 0 retweets 0 likes -
Semantics. I could easily argue that none of CS is science. Point is, advancement is not mere brute force compute power.
1 reply 0 retweets 2 likes
OK, I grant the point… I would like to understand better how much of the advance is that vs brute force; that question seems mostly unasked.
-
-
A fine question. Good place to start is looking at work with new architectures that decrease parameters but increase accuracy/performance.
0 replies 0 retweets 2 likesThanks. Twitter will use this to make your timeline better. UndoUndo
-
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.