How was the HTM neuron discovered/invented?

I want to know how did you guys came up with this specific model of HTM neuron ?

Do you have a specific book/s or paper/s that made the decision more clearer it has to be this way ?

f.e. why 2048 bits, why 2% sparsity, why proximal vs distal, why ORing distal syn, why permanence … etc ??

Do you miss some functionality you excluded so that model can be computationally tractable ?

1 Like

The idea was derived from a theoretical need for neural tissue to have two representations of input. One representation is the feature that a sensor senses and the other is a representation of that feature that is unique in context. Almost all the components of the model were known to exist. The insight was how those components could be arranged to achieve the needed result. This idea, plus relevant citations, are in the 2016 “Why do neurons have thousands of synapses” paper.


thanks … i have to reread your papers every 6 months it seems :wink: … always forget something … same with the videos … recently re-watching the sense-location-videos I have missed so many subtleties …

Now i was interested about “segments” and caught the detail (and updated the cheat sheat)… earlier it was all over my head :wink:

BTW the paper seems to be missing the Tables !

Those are in the Supplementary Materials. Their UI is not very obvious, but if you go to and then select “Download Article” it will give you the option to download those separately.