Numenta On Intelligence Podcast

media

#1

We are working on a new podcast. Here is the trailer:

First episodes will be a 2-part interview with Jeff, stay tuned later in July!


#2

This is great news!
Are you going to make it available on the popular platforms?


#3

Yes:

Working on Spotify…


#4

The first episode (parts 1&2) are available now!

This is kindof a sneak peek, since we will not be promoting them until tomorrow. :slight_smile: I hope you all enjoy this conversation!

If you have any specific questions about the content of this first episode, please start a new thread on the topic on #htm-theory.


#5

The Thousand Rats Model of Intelligence? :sunglasses:
Thanks for creating and sharing these. Always good to have dialogue and focused Q&A to understand the ideas and hypotheses promoted in recent Event Talks and SlideShare Presentations.

If our models have consistently and historically missed 50% of how the brain is doing perception (e.g. the spatial / location signal / grid-cell like component) … aka … Rat figuring out how to get home … then the ‘Thousand Brains’ blog post from last year might have to be rewritten with a new title: The Thousand (or Billion) Rats (sensorimotor) Model of (figuring out how to navigate and path-integrate towards home) Intelligence.

Permit me the mid-week metaphor!


#6

It is actually thousands, one for each cortical column.


#7

Back of envelope calculation:
Column spacing ~= 0.00003 meter
Cortex area ~= 0.3 square meter
(0.3 / 0.00003)²
100,000,000 columns.


#8

Maybe. Truth is that we are not sure. I’ve seen conflicting reports.


#9

That sounds like the number of minicolumns. Cortical columns are roughly 0.1 to 1mm square. Minicolumns are spaced about 40-80 microns apart.


#10

Where is my go-to reference on the difference between columns and mini-columns with a definitive breakdown of the functional boundary for each.

For bonus points - where each fits in the Numenta models.


#11

The Columns paper describes this.


#12

Yes, I was using the 30–60 μm mini-column figure for my calculations.

From the referenced paper, “For example, cortical columns vary from 300 μm to 600 μm in diameter (Mountcastle, 1997).” This puts me off by a power of 10 for the 0.00003 meter figure that I used.
Revising with this corrected figure:
Column spacing ~= 0.0003 meter
Cortex area ~= 0.3 square meter
(0.3 / 0.0003)²
1,000,000 columns.


#13

Oh. There it is.
The Million Rats Model of Intelligence. :sunglasses:
thanks for the good humour as well.


#14

How deep the minicolumns stretch matters too. They might just be around 5 cells per minicolumn in L5. I can’t access it anymore so I might be remembering wrong: http://science.sciencemag.org/content/358/6363/610


#15

Just finished part 2 of the podcast and watched Jeff’s recent talk on Grid cells, this is really fascinating stuff.

When I think about how the brain naturally maps the environment, it really does seem to be based on reference points between objects.

If you can imagine yourself going for a long walk, you’re not thinking of the entire path at once, you think of it in parts. For example, the path leaving from your house to the side walk, from the side walk to an adjacent street, from the adjacent street to the park entrance, etc.