Maybe progress in Vision/NLP/etc is not due to neural nets, but rather due to the influx of smart researchers, lots of compute, and industry pushes. Focusing on a different paradigm (graphical models, evolution, ..) could’ve made the same amount, but different type, of progress.
-
It’s like Evolution Strategies, but for AI techniques. We didn’t optimize for diversity so all that survived are the neural models… Probably not the optimal outcome.
The field of DL optimizes neither for diversity nor usefulness, it optimizes for hacks that are hard to evaluate & reproduce, and thus easy to publish. Bonus if there's some sci-fi terminology or shiny pictures involved that can be used for PR.
-
-
I agree that this kind of work makes up a large fraction of what’s currently being done DL (at least NLP from what I’ve seen). But there have also been a few real breakthroughs, though they may also have happened within a different paradigm, and without neural nets.
-
There are undoubtedly plenty of breakthroughs. But the signal/noise ratio is just not very good. I hope we'll do better as the field matures.
- Show replies
New conversation -
-
-
Cool
Thanks. 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.