Should be bursting complexity considered in more detail in HTM?

Hi All,

Bursting seems to be complex and heterogeneous across the pyramidal cells in the cortical micro-column. Really, as physical phenomenon, is fascinating [1].

The intuition of HTM seems to fit quite nicely (i think). My understanding is that if a cell soma reach the threshold tension without enough depolarisation in the distal tree, the spike is reflected into the soma again. This reflection might generate a burst through the axon (with certain probability). The bursting probability seems to be network-depentent status forming and hysteresis loop [2]! (beware, are model derived numbers). [If you to something like that in the current HTM algorithm (by assigning a network state dependent probability to the dendritic segment creation), still works!].

What intrigues me is that there are at least two different bursting pyramidal cells in the column (with clear different behaviour): FRB in supra granular layers and IB in infra granular layers. The current sensory-motor models are not considering this (the algorithm if the TM is mainly the same in all layers). It looks like a control circuit to me.

Any thoughts about this? Perhaps is not important … perhaps it’s… just wanted to share some big doubts about this apparently small details.

[1] http://www.scholarpedia.org/article/Bursting
[2] https://www.nature.com/articles/srep26029/figures/6

2 Likes

It might be important. Our attitude at Numenta has been to get as far as we can with the theory without incorporating oscillations and bursts, knowing that we might need to incorporate these concepts eventually.

1 Like