This meeting will be mostly remote because of COVID-19. We’ll see how it goes. @FFiebig will be recapping his attendance at COSYNE. I’ll be streaming my view. Live today at 10:15AM PDT.
Here is Flow’s presentation.
The link is ifixed!
Wednesday’s meeting will be live here:
We should be live within 30 minutes.
Marcus mentioned this paper:
Trying to understand how do you guys make sense of the information you got from COSYNE and add in your “box of knowledge”? Do you work with MindMaps to track things, other techniques? How do you try to connect the dots, please? Any advice, please?
I found need for enough place fields near walls to resolve the line to not cross. Otherwise between places are low field strength areas that resemble openings, wall is closer than appears.
For the inner navigable area the place fields have to be made as large as possible, or unnecessarily slows propagation time across clear areas to run through. Very large place fields would benefit from the mentioned clustering around rewards.
The arena used in the paper was small enough for the place fields to be equal size for both. And now I know how to scale up the arena size way up without associated performance loss!
Varying field size complicates the math, but I have new vesicle maker code. Before adding to my topic for that project I could try varying the field size at each point according to distance from solid objects and attractor/reward starting traveling waves outward in all directions, mapped somewhere on the sphere.
The rat video looks to me like a motor system keeping up with the navigation network controlling where its mouth goes, while alternating between spatial frames similar to “room” and “arena” in Dynamic Grouping of Hippocampal Neural Activity During Cognitive Control of Two Spatial Frames.
That was another awesome research meeting. Thanks!
While working on the new vesicle maker code I thought about the way the Blue Brain Project video shows many more cells than necessary propagating a traveling wave. One cell is enough to span from column to column with a single action potential event, instead of gradually increasing wave front of activity.
The answer seems to be to connect further out than (as now) nearest neighboring (same type) cell or column, so others in between can save their energy for when needed.
I was also thinking that theta alternating of activity, with another, may from the perspective of a neuron be a way to get a needed break so they can do something else, maybe while quiet check their dendritic inputs for what next to do? Having inputs from different sensory areas adds other detail to the picture. For at least the navigational model I work on these subpopulations would be sharing information they detected, by passing, not passing or reflecting waves according to properties of what is at that place in map.
The new vesicle code will still have equal sized cells or columns, with existing pulse wave for verifying that signals will cleanly go all the around then always cancel out on the opposite side.
For 1000 points and larger vesicles I expected to be able to get an even distribution of 5 sided pores, but always get fault lines that divides the sphere into a number of interconnected areas, with what are best described as folds between still partially connected tectonic plates where perfect hexagonality makes them hold together as one unit. After removing the spherical buldge from the picture, areas in between are geometrically flat 2D hexagonal arrays.
To get the plates to quickly form I have been randomly changing radius, as though experiencing pressure changes as in shaking in a jar of water, or caught inside pounding shoreline waves. After stopping the churning it looks like (without any parallel processing) it will take the current 8192 point vesicle another hour to relax into even better symmetry. By using that twice at two angles to start off a 16384 cell (or more) example I should by tomorrow have enough area for connecting further out than 1 neighbor to be required.
Tomorrow 10:15AM PST.
In part 2 Jeff explains an idea that grid cells might be misinterpreted and can essentially be place cells within environments of a particular scale in which there will be no repetitions among the cells in a given module. If so than, given two grid cell modules of different scale, can cells from a smaller scale one be interpreted as place cells within a single firing field of a cell from a larger scale one? An animal could use those when it needs to make a higher-prescision movement.
Also on the video at the very begining of part 2, the oval that goes from rodent’s head changes not only in direction, but also in length (scale?). Not sure what exactly this oval represents though
And another question: is there any correlation between place field length in a given direction and an average speed of a rodent in that direction within this place field?
@FFiebig - you covered a poster ( II-107 Hexadirectional coding of decision trajectories through abstract and discrete spaces. Seongmin Park, Douglas Miller, Erie Boorman) that talked about hex-coding in the major hub areas. Do you have papers that go into this in more detail?
This would go a long way towards proving my strong hunch that hex-coding is the lingua franca for all the hub areas in the cortex.
@FFiebig - the spatial nature of “grid clls” - how much of this is due to testing with spatial tests?
You brushed on other stimuli that formed grid type response.
If the test was social interaction would we have greet, dominate, sniff, etc cells?
@FFiebig Covert attention task discussion (talk # 3) - the visual field has many objects in it. I don’t see any reason to assume that the only things that show up in the HC/EC blackboard would only be certain “important” objects. Unimportant objects may become important. I would hazard a guess that the “noise” that you see in these studies is from the fact that there are many objects outside of the ones that the experimenters feel are the focus of the experiment.
The question I have here is how fast this blackboard is updated.
I could speculate that the blackboard gets refreshed extremely fast each time my attention jumps to another object. And so in essence I only have attention for a single object or even a single feature of an object at a time. And the speed at which I can shift my attention gives me the impression (dare I say illusion?) of observing many objects and features in a scene.
I think this is part of the interesting binding problem or the multisensory integration problem.
See part two at time index 8:00. watch the interplay with the room grid and the head cells - it’s really fast.
See part three, time index 6:00 - the black board has components that are NOT the focus of attention.
I thought it is attention, but decoupled from eye movement.
But I’d think this would also happen with imagined locations.
I dont know, but I noticed the red grid pattern is denser or maybe more in focus when the white loop is more elongated.
I think this loop is a calculation based on the inputs from the cells and the pattern is another representation of that same calculation.