Dynamic compartmentalization in neurons enables branch-specific learning

Good paper strong on compartmentalization of the dendritic tree.


Thanks - looks v interesting!

Branch specific dendritic learning is required in our models, and it’s interesting to read about the various ways this can happen in neurons.

This paper presents a dendrite model with dynamic dendrite compartments which respond to inhibition and allow for cooperation between different areas of the dendrite. They formalize the concept of a compartment in their electrical model of a dendrite.

I simplified the papers dendrite model and incorporated it into the existing Nupic HTM. I then tested it on the Numenta Anomaly Benchmark where it performs as well as the prior Nupic HTM model with static dendrite compartments.

My model allows for the HTM dendrites to be modulated the ways which are described by the paper, although this is not currently part of my model.

My write up is at: