To be clear, we may at some point realize that credit assignment is not key. But at this point, that's total pie-in-the-sky. It's not reasonable for scientists to reject an area of study because of an unsubstantiated belief that some future discovery will render it obsolete.
-
-
Replying to @tyrell_turing @pfau and
Stupidly late to these comments but very well articulated, Blake.
1 reply 0 retweets 2 likes -
Replying to @SussilloDavid @tyrell_turing and
I think it's perfectly reasonable for a scientist to choose to avoid a certain field of theoretical research because they don't believe the tools are there to falsify it experimentally.
1 reply 0 retweets 6 likes -
Replying to @pfau @SussilloDavid and
Yes, totally. But, that's precisely what we have to change (and what a few groups are working on, including mine): our credit assignment models need to start making physiological predictions that can be falsified.
2 replies 0 retweets 6 likes -
Replying to @tyrell_turing @pfau and
Wait, credit assignment = backprop?
2 replies 0 retweets 1 like -
Replying to @xaqlab @tyrell_turing and
The brain does credit assignment by whatever was cool at NeurIPS 5 years ago.
1 reply 0 retweets 3 likes -
Let’s add meat to this then, though. What are good options for credit assignment that *don’t* require efficient access to an estimate of the 1st order gradient of an objective function w.r.t. a given synaptic weight deep in a network? Honest question.
1 reply 1 retweet 3 likes -
Replying to @AdamMarblestone @pfau and
(I realize many answers will restrict the architecture or objective fxn greatly, to allow specialized non-backprop ways to get such gradients, which wouldn’t work for general fxn approx in arbitrary net topology. What can you really do with those? What’s the best alternative?)
2 replies 0 retweets 1 like -
Replying to @AdamMarblestone @pfau and
I guess it breaks into 2 cases: 1) you don’t need the gradient, or 2) you do but you have a way to get it that is structurally very different than backprop. For instance this paper gets gradient w/ either backprop or EM (perhaps an example of case #2): https://arxiv.org/abs/1202.3732
1 reply 0 retweets 3 likes -
Replying to @AdamMarblestone @xaqlab and
Also I think "the brain minimizes an objective function" is a vacuous statement. All dynamical systems can be reframed as solving a variational problem.
3 replies 0 retweets 5 likes
Yes, I think a part of this framework is that it would be a useful objective fxn that the brain would actually *compute* and *learn from online*, and perhaps could be tweaked in some direct, natural way by evolution. What we would otherwise call an error signal basically.
-
-
Replying to @AdamMarblestone @pfau and
I seriously want to know more though about neural learning systems that are very powerful but where you either don’t need the gradient w.r.t. weights or have a very non-backprop-like way to get it due to some restriction of the architecture.
0 replies 0 retweets 1 likeThanks. 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.