Memory consolidation by Hippocampus as combination of old memories with new, abstract and generative?


Hmmm, interesting paper. However, per the HTM roadmap, I think these elements of intelligence come in the way of truly understanding sparsity and hierarchy. Very briefly- the nature of hippocampal replay is obviously very nebulous, but we can all agree that its not doing something as fundamental, and ground-up as the functions outlined in the paper. Invariance and hierarchy are embedded in the functions of the network itself, and cannot be relegated to mechanisms that reside outside of it. There are some recent paper adjoining sparse architectures to back propagation and how such combination dramatically reduces catastrophic forgetting…


I think the difference is in degree, REM sleep allows for much longer generative phase in HC.
I agree that there should similar mechanism in cortex, probably in thalamo-cortical loops, but day/night cycle is very prominent in brain functioning.

1 Like