Modelling of Distal Dendrites and Synapses

Hi. I’ve read the new temporal memory white paper [1] a few times but still don’t fully understand the modelling of the distal synapses.

  1. Is the number of “potential” distal synapses per dendrite segment static (defined as a constant or as a percentage of all cells) or dynamic (i.e. a segment “starts” with x synapses but then dynamically adds/grows new synapses?)

  2. How are the target cells selected that the distal synapses link to? I.e. do these target cells have to be in a “radius” or “neighborhood” of the respective cell or can their location be chosen randomly? My point here is, whether – based on current theory – location/distance of laterally-linked cells matters.

For the proximal dendrite HTM clearly defines the number of potential synapses (=50% of input bits) and their selection process (=random, unique set). But for distal synapses the model is less clear. Would be great if somebody could help clarifying above points.

Thanks in advance.

Matt

[1] http://numenta.com/assets/pdf/biological-and-machine-intelligence/0.4/BaMI-Temporal-Memory.pdf

2 Likes