@niftierideology @voidfraction @admittedlyhuman > winning local configurations; which isn’t interesting, is it?
-
-
Replying to @Meaningness
@niftierideology@voidfraction@admittedlyhuman Not interesting inasmuch as NN not required. Maybe much simpler method would have worked.1 reply 0 retweets 1 like -
Replying to @Meaningness
@niftierideology@voidfraction@admittedlyhuman Missing control experiment: for same amount of GPU time, would simpler ML method work better2 replies 0 retweets 0 likes -
Replying to @Meaningness
@Meaningness@voidfraction@admittedlyhuman I'd be interested in the outcome of that experiment1 reply 0 retweets 0 likes -
Replying to @niftierideology
@niftierideology@voidfraction@admittedlyhuman I am seriously interested in doing that for the image recognition stuff.1 reply 0 retweets 0 likes -
Replying to @Meaningness
@niftierideology@voidfraction@admittedlyhuman For AlphaGo, I think the likelihood of finding it did something interesting is too low.1 reply 0 retweets 0 likes -
Replying to @Meaningness
@niftierideology@voidfraction@admittedlyhuman And the compute cost is so high that no one is going to challenge Google’s results.1 reply 0 retweets 0 likes -
Replying to @Meaningness
@Meaningness@voidfraction@admittedlyhuman I'd be kind of surprised if the supposed state-of-the-art method was actually superfluous2 replies 0 retweets 0 likes -
Replying to @niftierideology
@niftierideology@voidfraction@admittedlyhuman Well, I found that, repeatedly, when analyzing NN results around 1990.1 reply 0 retweets 0 likes -
Replying to @Meaningness
@Meaningness@voidfraction@admittedlyhuman Maybe it's easier to use a NN when a more custom-fit solution would take longer to discover?1 reply 0 retweets 0 likes
@niftierideology @voidfraction @admittedlyhuman Yes, as an engineering matter, in current state of the art, this is often right.
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.