shouldn't this be very easy to reimplement though?
-
-
Indeed, and I will gladly eat my words if it turns out to be completely reproducible.
2 replies 0 retweets 0 likes -
But if I had a bombshell like this, I wouldn't have sit on it for 8 weeks of NIPS reviewing; makes me suspicious.
1 reply 0 retweets 1 like -
The result isn't that surprising is it? Roughly stacked sparse coding. Similar models have been built before
1 reply 0 retweets 0 likes -
Only difference is claim to match VGGnet. But sparse-coding has given strong numbers in past
1 reply 0 retweets 0 likes -
Layer-local greedy training doing that well is pretty surprising, yes.
2 replies 0 retweets 3 likes -
the reason DL works so well is the joint training of the stacked reprs. Greedy approaches fundamentally flawed
1 reply 0 retweets 5 likes -
Maybe. It doesn't seem logical that joint training should ever lose to greedy training, but if it does?
2 replies 0 retweets 0 likes -
I'll just say "prevailing wisdom" in DL is a fickle thing. You've seen the silly shit we all believed in 2010.
1 reply 0 retweets 4 likes -
agreed, but at the same time I've seen some silly shit accepted at major conferences. Jury's out on this one
2 replies 0 retweets 0 likes
for the record I do believe backprop will eventually be replaced; I've said this many times before.
-
-
our brain learns pretty well and probably doesn't backprop
2 replies 0 retweets 0 likes -
Depending who you talk to :) Geoff has long said that error signals are just too damned useful...
2 replies 0 retweets 1 like - Show 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.