Use TD Learning (a real cortical algorithm?) and Hebbian Learning with NuPIC

neuroscience

#1

HTM Theory is trying to implement the algorithm in our brain. Can we see some of these things in HTM Theory/NuPIC that we observe the neocortex to be doing? Namely, Hebbian Learning and TD Learning?

very nice notes about classical conditioning in the human brain and its computational aspects: https://www.princeton.edu/~yael/PSY338/5%20Classical%20Conditioning%20IV%20online.pdf
https://www.princeton.edu/~yael/PSY338/6%20Classical%20Conditioning%20V%20online.pdf
https://www.princeton.edu/~yael/PSY338/4%20Classical%20Conditioning%20III%20online.pdf

TD Learning has been observed in the brain:

It might be the actual cortical algorithm that Jeff Hawkins talked about in his book, since it’s literally the algorithm used for predictive learning in the brain.

TD Learning in the brain uses dopamine. Humans have much more dopamine than other apes which potentially lead to our intelligence.


#2

Thanks for asking.
I’m not familiar with RL in HTM (Not many has implemented it). But I’m pretty sure SpatialPooler and TemporalMemory is Hebbian Learning.


#3

You should read this thread: HTM Based Autonomous Agent

It combines Hebbian learning & TD learning.