Neural networks are not a fad this time. There will be new things invented, but many of the fundamentals are here to stay for a long time. Happy to put money on this. Optimization and regression aren't fads and neural networks are just simple, good, and easily optimizable models.
-
-
Replying to @DavidSHolz
Yes, they are _just_ that, inference of how the brain works from them is misguided!
1 reply 0 retweets 0 likes -
Replying to @posobin
We can agree to disagree on this. They certainly feel like useful models to me. At least like how classical mechanics is useful for approximating macroscopic phenomena.
2 replies 0 retweets 0 likes -
Replying to @DavidSHolz @posobin
David is right. Gleb: https://www.frontiersin.org/articles/10.3389/fncom.2016.00094/full … There is no such article one can write for SVMs
1 reply 0 retweets 2 likes -
Thanks Jose! I know there are a lot of great articles on this topic, but I'm in a cafe and can't recall them off the top of my head. You're so encyclopedic!
Are there a bigger set of articles beyond this one that we should refer to?1 reply 0 retweets 2 likes -
-
You could start with this: https://docs.google.com/document/d/1qil2ylAnw6XrHPymYjKKYNDJn2qZQYA_Qg2_ijl-MaQ … And this: https://www.nature.com/articles/s41593-019-0520-2 … And the hundreds of refs in my paper above...
1 reply 0 retweets 2 likes -
Replying to @AdamMarblestone @ArtirKel and
But it is just a hypothesis still, as regards deep learning per se. One should also consider attractor based neuro-symbolic systems, say, https://arxiv.org/abs/1611.03698v1 … and structured PGMs, when thinking about possible frameworks for brain computation https://arxiv.org/abs/1808.01058
1 reply 0 retweets 1 like -
Replying to @AdamMarblestone @ArtirKel and
And of course it doesn’t mean the brain is just an image classifier or something. Would look at integrated agent models like this: https://arxiv.org/abs/1803.10760 We need real neuroscience to find out what is really going on though!
1 reply 0 retweets 1 like -
Replying to @AdamMarblestone @ArtirKel and
This is notoriously hard as a twitter discussion but hopefully some of those papers will help!
1 reply 0 retweets 0 likes
What does it look like to try to do the same for a symbolic approach? Perhaps production system models like ACT-R http://act-r.psy.cmu.edu/wordpress/wp-content/uploads/2012/12/526FSQUERY.pdf … or AI visions like https://groups.csail.mit.edu/genesis/
-
-
Replying to @AdamMarblestone @ArtirKel and
I like Eliasmith’s modeling work which is kinda deep learning agnostic https://www.amazon.co.uk/How-Build-Brain-Architecture-Architectures/dp/0190262125 …
1 reply 0 retweets 1 like -
- 2 more 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.