Numenta Research Meeting - Nov 4, 2019

Jeff will probably talk shortly about Hierarchical organization of cortical and thalamic connectivity. Not sure about other topics.

Today at 10:15AM Pacific.


The Murray Sherman talk from Berkeley 2018.


@htm-cadre This is an interesting research meeting to watch if you are trying to understand that cortical - thalamic “loop” (Jeff says Murray Sherman says it’s not a loop).


This paper also addresses the interplay between cortex and thalamus:

Pro tip - kind of a hard read, but well worth the effort!


I just missed it again, but I watched the rerun.

Is there more information about the multiplexor idea? @jhawkins mentioned a complex circuitry that looks like it can select or rerout connections. That looks like a great research talk on itself. I’m wondering (but very speculatively and probably noobishly naive) whether this multiplexor could select several outputs at once. @subutai said that there are ten times more connection one way than the other. Is each of those 10x connections exclusive or could it work like some kind of additive encoding bus where each line represents some order of magnitude. (So for example if lines 1, 2 and 5 are active only, the output codes for 10011). This could be the stearing signal for a size multiplyer. (Even stochastically with lots of redundancy this might work I think).

Also @FFiebig mentioned the oscilation rate between circuit centers. It’s not the first time he talked about that. This sounds intriguing. I’d love to hear more about that.

Great stream again. I feel so fortunate to be able to witness all this. Thanks Numenta!


I’m a cortex-only kind of a guy (besides a little hippocampal work in the first year of my PhD), so this is all terribly new and exciting to me as well. Plenty of work to read now. With regards to oscillations, I might add that I’ve probably payed a lot of attention to supragranular gamma because the microcircuit models I use (which rely on the provably fast and efficient feed-back inhibition from paravalbumin-positive, fast-spiking, basket cells) are very good at explaining recorded data. Infragranular layers (5,6) typically do not reach 80+Hz gamma bursts. There are models that predict this (see the work of Andre Bastos and Pascal Fries in particular) but I think Jeff was talking about other oscillations, like 30Hz upper beta. I’m hoping to do some comparison of predicted activity levels and oscillations between cortex models before Christmas rolls around, so we’ll talk about this in more depth eventually.


Papers which mention L4 corticocortical projections:

  1. A genuine layer 4 in motor cortex with prototypical synaptic circuit connectivity
    They found sparse cells projecting from M1 to contralateral M1, as well as M2, S2, and/or S1 (i.e. retrograde tracer was injected into all three of those regions in the same animal.) There were no spiny stellate cells.

  2. Excitatory Local Connections of Superficial Neurons in Rat Auditory Cortex
    This paper argues that L4 in rat A1 is similar to L2/3. For example, L2/3 basically doesn’t project into L4 in V1 and S1, but it does in A1. A1 also lacks spiny stellate cells. A1’s L4 has dense corticocortical cells.

  3. Channelrhodopsin-2–assisted circuit mapping of long-range callosal projections
    The paper cites that callosum-projecting cells are in all layers but mostly L3 and L5. I can’t access their cited articles (their refs 26 & 27) and both are about development.

1 Like