I think it's as useful a proposal as any out there....but that's just it, needs empirical data to support. I'm no computational expert...but are we really at the point where we think AI solutions have transcended wet brains? (still have to read the paper in detail).
-
-
Replying to @JasonSynaptic @schoppik and
No, but that's not the argument. The core argument is that the computations in the brain are the result of two optimization processes, and these are not necessarily going to produce computations that we humans can easily explain, just as ANNs are "uninterpretable".
3 replies 0 retweets 9 likes -
Replying to @tyrell_turing @JasonSynaptic and
As such, we need to frame our questions about computations in the brain using the things that really guide the optimization processes, namely, architecture, objective functions, and learning rules.
1 reply 0 retweets 1 like -
Replying to @tyrell_turing @schoppik and
But these are all artificial constructs that work outside a brain...but what evidence is there that this *could* be how real neural networks work?
4 replies 0 retweets 8 likes -
Replying to @JasonSynaptic @tyrell_turing and
Evidence? Learning rules are what synaptic physiologists study. Architectures are what circuit physiologists and EM/connectomics people study. Objectives are what behavioral and evolutionary biologists study.
2 replies 0 retweets 7 likes -
Replying to @neurosutras @tyrell_turing and
Ok cool. So we've solved the brain then?
1 reply 0 retweets 2 likes -
Replying to @JasonSynaptic @tyrell_turing and
No but these areas have essentially already been identified as important avenues to understand brain function
1 reply 0 retweets 2 likes -
Replying to @neurosutras @tyrell_turing and
I would hope so! But, you're making it sound like we just need to "plug'n play" some sort of AI algorithm using these different sets of analysis...and then we've understood the brain. I don't see how my Q is so far off base...
1 reply 0 retweets 1 like -
Replying to @JasonSynaptic @tyrell_turing and
Pretty much the proposal is, "keep doing what you were already doing, but maybe think about it differently and temper your expectations about what a real answer to a neuroscience question might look like." Not. That. Radical. And not suggesting replace neuro with deep learning.
2 replies 0 retweets 14 likes -
Replying to @neurosutras @JasonSynaptic and
Amen!
1 reply 0 retweets 0 likes
And find out if brain can *actually* optimize w/ credit assignment thru multiple layers... otherwise you’re stuck with *evolutionary* opt, which may look very different, e.g., dev. programs, re-usable circuit motifs & molecular wiring rules... strong form of this is empirical...
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.