Hex Grids & 1000 Brains Theory


#29

This feeds into the larger question of how a hex-grid is learned and how the learning extends.

As I see it a learning starts with some random best-fit on a single mini-column.

As others learn some pattern at the same time they riff off each other and reinforce when they see their part of a common pattern.

Over time this patch grows as a patch learns this pattern together. Once is it established it adds details around the edges and learns to discriminate between two similar patterns. The yellow in the picture below is a loose group - say 50 hexes. As the pattern is refined more hexes are recruited and by the time we get to the green it might be 500 hexes.

Look at this process in action: The two patterns start out the same but as details are added in the two patterns start to diverge and I would expect that the phase or scale shifts between the two patterns.


#30

oh… thank you I have to think about this now… so the atomic computation unit we see physiologically localized in sensory cortex is distributed in other parts of cortex… right?

So the cortical column is distributed into a learned hex grid in higher areas of cortex, each hex grid behaves like a grid cell module, bumps within the module are minicolumns firing / echoing from some recognized sensory input (or something else). Am I still on the right track?


#31

Yes, with the key difference being that in the early stages we are anchored to the sensory feed - the location have to be fixed as the sense fibers are fixed.

As we move up the hierarchy the hex-grids are free to form at any mini-column location and shift (phase/spacing/rotation) to collecting the sensory information into hex-grid coding of objects.

Then go back and look at the video you made on grid signalling and plug this in. I think you will see that it is describing the same thing from the bottom up.


#32

when you say “hexes” you mean hex grids working together yes? So I could also say cortical columns here.


#33

Yes - same thing - different terminology.

The curse of working in the AI field.

Please keep in mind that Calvin defined this a long time before HTM was offered as a thing. I have been thinking this way since the 90’s.

http://williamcalvin.com/bk9/index.htm


#34

Hexagonal topography has been around along time. Biology finds interesting ways to use maths.

I think I understand what you mean by “well shuffled” now. Well shuffled up the hierarchy.


#35

Yes - the “dendrites have thousands of synapses” but they have to have something to sample and they really don’t reach very far.

If we are going to extend this reach we have two mechanisms:

  • Bring more mixed input to the dendrite to sample
  • Harness them together across some expanse with a system like hex-grids.

I was ready to see SDRs as the best input coding scheme; it has much to offer.


#36

btw, do you mind if I update your post to say “minicolumns” where it should instead of “columns”? HTM Columns into Hexagonal Grids!


#37

Go ahead - it’s really the right thing to say.

gmirey pointed out that I was very sloppy with my terminology and that it was very confusing for him.

If I ever get everything down that I am thinking I will end up collecting this up into book-length material and publishing it in some form. Editing to get everything to match up will be a large task.

So much left to get down.


#40

I like the idea that different CAN parameters can be used to extract different hex grids from a set of activated minicolumns. I don’t think we are considering representation at this level. We are only using overlap score with input within columnar boundaries.

I have yet to talk to Marcus about it (this afternoon), but the idea of doing a hex-grid search across a set of activated minicolumns with proper topology, getting a ranked list of grids and inspecting what that means is quite interesting.

I talked to Jeff about it this morning and he still thinks that the lateral voting as defined in the Columns+ paper does the same thing. I was directed to read and understand our voting mechanism better. Admittedly, this is not something I’ve done a video on yet so I don’t fully understand it.


#41

I suspect that Numenta has not considered that approximately fixed length interconnections will naturally form a triangle, and from that eventually, a hex-grid.

I discussed the possibility of interlocking grids in my post and expect that this is or is not possible based on how strong and long-range the inter-neuron inhibition is.

the idea of doing a hex-grid search across a set of activated minicolumns with proper topology, getting a ranked list of grids and inspecting what that means is quite interesting.

I see that you are starting to see the possibilities that this opens up for inter-area signaling.

You may also think about what it might mean if more than on one grid forms on a map at the same time and how that is related to some of the papers that have been describing what they are seeing with coding in the HC/EC complex. At a bare minimum, this is what is in the cortical hubs. (Association regions)

I think about this a lot; tuples come up a lot in my thoughts.

Keep me posted.


#42

Mark, I spent some time with Marcus and showed him how a set of minicolumn activations could be resolved to hex grids using the techniques you described. I think he fully understood the mechanism; he agreed that it was very simple, but he also did not know why it would be needed or what would be achieved with it. He considered it an interesting trick that could be used, among many other interested tricks biology uses to do interesting things. He did not seem to think it was the source of any big ideas.

I still need to understand how column voting works better, so I will be re-reading the Columns paper. I will continue thinking about hex-grids as I progress through the material.

I’m sorry nothing has come of this line of thinking at this point. If I see anyone using this trick in any way here at Numenta, I will be sure to attribute it to @bitking / Calvin.


#43

No harm - no foul.

On the plus side - you have confirmation that it should work from someone you trust.

When they get around to trying to connect maps together I think they will get religious on this.
Until you need to signal from one area to another there is only one local advantage - clean sparsification without the k-means thing.

If you can point me to which paper is considered the Columns one I will also read it again. Perhaps I can add something to what they think they know. I assume it’s this one:


#44

That’s the one. I am afraid you’re going to have to get into the formulas in the “methods” section at the end if you want to really understand the lateral connectivity in the model. That’s where I’m at.


#45

If you read the columns paper assuming that the L2/3 is doing hex-grids fits very nicely in the descriptive text and in some ways, suggests hex-grid behavior - for example: while L2/3 and L5 cells exhibit “complex” RFs (Hubel and Wiesel, 1962; Gilbert, 1977). Key properties of complex cells include RFs influenced by a wider area of sensory input and increased temporal stability (Movshon et al., 1978).

and

Cells which have similar classic receptive fields when presented with isolated edge-like features, diverge, and fire uniquely when the feature is part of a larger object.

…snip…

To explain border ownership, researchers have proposed a layer of cells that perform “grouping” of inputs. The grouping cells are stable over time (Craft et al., 2007).

  • I am certain that the function of L4 works as a timing coordinator with the interface with the thalamus for the formation of waves and local synchronization to these waves. I have supporting papers on this but that that is not the issue I am working now.

I predict that the timing below will be related to the gamma rate (40 Hz) or 25 ms:

Activations in the output layer do not require very fast inhibition. Instead, a broad inhibition within the layer is needed to maintain the sparsity of activation patterns. Experiment evidence for both fast and broad inhibition have been reported in the literature (Helmstaedter et al., 2009; Meyer et al., 2011).
Our simulations do not model inhibitory neurons as individual cells. The functions of inhibitory neurons are encoded in the activation rules of the model. A more detailed mapping to specific inhibitory neuron types is an area for future research.

Wait - what was that last bit? …
The functions of inhibitory neurons are encoded in the activation rules of the model. A more detailed mapping to specific inhibitory neuron types is an area for future research.

The biology seems to point to smaller local pools of inhibition that are triggered locally. This is not rocket science - a little stroll through related papers should lay out the scope of inter-neurons receptive fields and modulating output connections.

While I am banging on non-biological problems with this paper here is a monster one:
In this experiment, each column receives lateral input from every other column.
This is absolutely NOT how it works in biology.

These connections are the foundation of the Hex-grid formation and the model skips right over it.

What does the biology say?
https://www.researchgate.net/publication/12675879_Horizontal_Synaptic_Connections_in_Monkey_Prefrontal_Cortex_An_In_Vitro_Electrophysiological_Study
From this paper in my hex-grid post:
What Proportion of Layer 3 Pyramidal Cells Receive Long-distance, Excitatory, Monosynaptic Inputs?
Our findings also suggest that most pyramidal neurons in layer 3 are targets of long-distance, horizontal projections. Specifically, low-intensity stimulation at long distances from the recorded layer 3 pyramidal cell evoked monosynaptic EPSCs in the majority (77%) of these neurons. However, this proportion is likely to be an underestimate since some long-distance axon collaterals were probably severed by slicing of the tissue blocks. Although the present study does not indicate whether horizontal projections synapse selectively onto layer 3 pyramidal neurons, our results show that these cells frequently receive this type of synaptic input

I have several more paper (including ones referenced by Numenta) that support the exclusively longer length of lateral connections. When you consider this go back to my third drawing showing a “halo” around a given mini-column in the post above.

And then consider how that drives the formation of triangular formations.

So the answer to hex-grids questions was “we considered that - go read the columns paper” and “sure it would work but why would we look at that?”

  • How about - because the biology does it?

More comments on the columns paper:

Testable Predictions
Hex-Grids are strongly compatible with most of these with no changes.

Methods section
Ok, I read the methods section again - it is what I saw the first time I worked through the math.
Here’s where they play footloose and fancy-free with the biology - the tangle of axons that fan out from the mini-column is what activates the inhibition cells - not some vague inhibition field in the model.

Look very hard at this related bit in Computing Cell States

As I said above: this section totally ignores the known topology of interaction between the mutual reciprocal connections and the range and activation properties of inhibition inter-neurons. When it comes to interactions between neurons I can’t stress this enough: Topology Matters. As offered in this paper the model continues the earlier simple “thousands of synapses” models and ignores the actual biology.

As described, the calculations based on this model is correct. The problem is that the model is NOT based on the known topology of the biology so it is misleading - it does NOT serve to explain the behavior of the biology. Numenta claims to be biologically inspired to explain how the brain works; they will have to do better.

The hex-grid proposal faithfully models this aspect of the biology.

I would start with modifying the model so that lateral connections are all at a biologically plausible distance from mini-column to mini-column. Your suggestion about looking at the hex-patterns that form should produce some very interesting behaviors.

I agree that Numenta has identified the inhibition function for future consideration. Why this is important is the bit I have mentioned before - the tuning of the ratio of distances of the mutual connections and inhibition range modifies the behavior of the column. At one ratio the column acts a Gabor filter - exactly what is needed for early sensory processing. Reduce the inhibition range and the column acts the hub of a hex-grid formation. I would think that this range of behaviors would be very important to the studies of cortical column theory.


#46

He had not considered the idea until I presented it to him yesterday. Once he understood it, he said it was not necessary for sensor fusion.

He understood how hex grids might emerge across minicolumn activations via CAN / mexican hat. He did not see the utility for our current work.


It pains me that you are disappointed, but this is really about Numenta working on this problem from a completely different vector for the past 10 years than you’ve been. We have been in the weeds, inside a cortical column, inside a layer, inside a minicolumn, figuring out detailed interneuron mechanisms to answer some core neuroscience questions and get a foothold into this common algorithm. That’s where we started and that’s where we still are. The ideas you bring to the table are certainly worthy of thought, discussion, scrutiny, and investigation, but they just don’t cross the direction we are heading.

I want to point out that both Jeff and Marcus saw the mechanism and understood it (much faster than I did). And neither denied that it was interesting and could be playing a role in computation at some level. But it is simply not overlapping with the questions we are currently asking about object representation.


#47

Please go back to your grid video and watch it with your new-found understanding of hex-grid coding.
Really, I know you made it but, watch it with your new eyes.
Then come back and tell me that there is no overlap with what Numenta is doing.

In particular, how else with HTM make phase/rotation/scaling the way hex-grids does it.


#48

Grid cell modules in entorhinal cortex and these hex grids are similar, but not the same. They both express the same characteristics, but their mechanisms are very different. Grid cells in entorhinal cortex do not exhibit the physical topology (hex grid bumps) that you describe. They form grids, but the grids are not physical hexagonal shapes formed by the placement of cells in the cortical substrate. Their bumps only exist in the space being represented.

We will use the grid cell tricks we can already see happening.


#49

I will keep slogging along on the track I am on, and Numenta, on it’s track.
I assume that they will cross over at some point.
I will keep up with the forum and help the nubies as I have been doing.
I am still onboard with HTM as described with the BAMI paper.
Were we diverge (Numenta and me) is the understanding of the distribution of object representation.

At this point I see Numenta looking with a laser focus on the single column and trying to explain as much of known behavior as possible with a local mechanism.

I have the different insight that this is a cooperative process between the various maps and in streams of connections. I see that this allows a much simpler local function and also see many connections to papers that I have read about this distribution of function.

Some of my foundation beliefs that drive this viewpoint:

  • The contribution of the subcortical structures in selecting behavior. (My dumb boss/smart advisor model)
  • The three streams paper is the closest to what I think happens in training the hierarchy.
  • I also really like Randall O’Reilly’s take on the prediction model in L5; it compliments HTM.
  • That the web of map interconnections is critical to understanding individual map functions.
  • The basic Global Workspace mechanism in connecting the need state to the perceived state.
  • My model of consciousness based on cortical connections, primarily the Arcuate fasciculus
  • The basic model of the hippocampus as a one-day buffer for personal experience
  • The contribution of subcortical structures in adding affective behavior guided by the amygdala.
  • That this affectively seasoned personal experience is transferred to the cortex in sleep
  • Most likely using spike timing learning.
  • That the basic HTM model is the best description of the local cortical computation and state transition.
  • That “thinking” is sequences of motor commands directed to maps.
  • That the cerebellum is important for guiding sequential actions inside the brain. (in addition to the usual motor actions driving the body)
  • That human speech is a learned motor action (generation AND perception)
  • That the learned motor programs of speech form many of the functions attributed to “higher mental actions.”

I see these as interlocking and forming a fairly complete model of human cognition.
When these are considered as a whole the local function of the cortical column is fairly well constrained


#50

I thank you. I’ve been working more closely with the research team lately, so I will be sure to keep an eye out for places where this particular bag of tricks could be applied. And if it does ever emerge from the research team, I’ll be sure to point out where it truly originated.