Wed 10/5 at 2pm: Deep Learning and Neuroscience

[Chad DeChant]

Hi everyone,

Our next paper reading and discussion will be Wednesday at 2 pm in the Computer Science Building conference room (next to the CS department lounge).

Jeff Seely will present the review paper, "Toward an Integration of Deep Learning and Neuroscience"  from Marblestone et al., focusing on a few of its key ideas. In particular, he will discuss some ideas related to biologically plausible learning, with a quick review of "feedback alignment" (from the paper, "Random feedback weights support learning in deep neural networks" from Lillicrap et al.). Jeff is a PhD student in the neuroscience program interested in how deep learning can guide theoretical neuroscience (and vice versa). 

Toward an Integration of Deep Learning and Neuroscience: http://journal.frontiersin.org/article/10.3389/fncom.2016.00094/full

Random feedback weights support learning in deep neural networks: https://arxiv.org/pdf/1411.0247.pdf

See you there!

Chad

PS If you have any questions or would like to take the lead in discussing a paper, please email me or Oscar (Oscar Chang <oc2...@columbia.edu>)

[Colin Raffel]

HI all, an interesting recent extension of the feedback alignment stuff will appear at NIPS this year: Direct Feedback Alignment Provides Learning in Deep Neural Networks.  In short, you don't even need to backpropagate the error signal between layers to get learning to happen, you just need to multiply the error at the output of the network by fixed random matrices.  Unfortunately the math in this paper is bogus/incorrect, but the experimental results have been verified.

∿

--
Schedule and info: http://labrosa.ee.columbia.edu/cuneuralnet/
---
You received this message because you are subscribed to the Google Groups "cu-neural-net" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cu-neural-net+unsubscribe@googlegroups.com.
To post to this group, send email to cu-neu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cu-neural-net/CAOrqLbCcdGiB6KmfTcg4zaq687OWHKTN6hVcyD6QTJ9byvu7Uw%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.