Numenta Research Meeting - June 22, 2020

Today’s research meeting covered 2 topics.

In the first part of the meeting Jeff discusses grid cells formed via oscillatory systems, the Bush & Burgess’ model of ring attractors, and how this idea might be overlaid onto cortical columns.

Then Subutai Ahmad reviewed a 2019 paper by Jeff Clune

“AI-GAs: AI-generating algorithms, an alternate paradigm for producing general artificial intelligence”


Video record (Youtube) of meeting? Sounds interesting.

1 Like

Starting at 34:00 Subutai switches gears quite a bit, and discusses a new paradigm for achieving AGI via meta-meta learning, by reviewing Jeff Clune’s 2019 paper “AI-GAs: AI-generating algorithms, an alternate paradigm for producing general artificial intelligence”


I have one question: Why is a baseline theta needed in order to get the model (any OI model) to work?

I understand, that we have a bunch of oscillators. When the animal moves in the preferred direction of the oscillator, then this oscillator speeds up slightly. For whatever reason, the oscillators never slow down. So if we think of a 1D case with 2 oscillators, then movement in one direction, makes one of the oscillators go faster. Movement in the other direction lets the other oscillator go faster. Interference occurs at the readout node (grid-cell).

But I do not understand, why a baseline frequency is needed. The model seems to work without it. Or is a baseline frequency just helpful, because it replaces the need for an extra oscillator, but the model works as well without the baseline, and another oscillator.

To have an interference pattern, you need two waves.

When both waves have the same frequency, the interference pattern moves with the origin (the animal). The same goes if there is only one wave.

When one of the waves’ frequency increases proportionally to the speed of the animal, the resulting interference pattern (with the unchanging base wave) becomes a traveling wave. If it is combined with interference patterns from several directional sensitive oscilators, the result becomes a 2D field that remains anchored to the environment the animal moves in.

At least, that’s how I understood it.

1 Like

Another question: Is there any evidence for theta oscillations in the neocortex and thalamus?

I looked at some papers and did not find much evidence for theta oscillations. E.g. Scholarpedia [1] groups theta into oscillations that occur during sleep. In my opinion they must be a prominent oscillation type, if one wants to attribute a central mechanism to them.
As oscillations are mainly enforced by inhibitory neurons, one could argue that they exist in the neocortex. But in the thalamus, there must be a cell type that can reliably exhibit theta oscillations and enforce them onto the relay cells. I could not find one.


I recall theta oscillations are clear in rodents but not primates. In the thalamus, there are inhibitory interneurons, though they’re sparse sometimes. There’s also the thalamic reticular nucleus, which receives input from relay cells and inhibits them. Relay cells themselves have burst / tonic mode which might contribute to oscillations.

1 Like