Re:Predictive Coding Approximates Backprop Along Arbitrary C...
by igorkraw on Tuesday October 06, 2020 @08:27AM
(#24696971)
Can we discuss via mail? You can find my contact in my profile. I started out with neuromorphic research before I switched to something else and this paper basically was my original research plan (but I never would have arrived at actually deriving this on the old trajectory I was on I think) so I'd love to go into the nitty gritty details
Re:Predictive Coding Approximates Backprop Along Arbitrary C...
by l33tman on Tuesday October 06, 2020 @10:31AM
(#24698021)
Yeah I'd be happy to chat some time, also been doing some PC-related work!
BTW the PC-stuff goes back to the 90's (at least), Rao and Ballard 1999 is usually referred to. The 2017 papers quoted above are specifically about comparing with backprop gradients.
1. James CR Whittington and Rafal Bogacz. An approximation of the error backpropagation algorithmin a predictive coding network with local hebbian synaptic plasticity.Neural computation, 29(5):1229–1262, 2017.
2. Rafal Bogacz. A tutorial on the free-energy framework for modelling perception and learning.Journalof mathematical psychology, 76:198–211, 2017.
(source: grad student working on PC networks. happy to chat if anyone wants)