Numenta Live Streams, 5 Feb 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Numenta Research Meeting - Feb 5, 2020

Description

Some discussion of Local Field Potential (LFP) from Florian, probably some random discussion of other things.

https://discourse.numenta.org/t/numenta-research-meeting-feb-5-2020/7147

A

B

A

Go quick today because it's mostly some bit of housekeeping for clarifying some other things we had in the discussion of.

A

A

A

A

Anyway, so the point was just to quickly break down an understanding of that, and could we switch to the other paper which is going to be the most important one? So I went through this review by which is probably the most authoritative review on a piece given that's been cited over 2,000 times in 2300 citations or something like that. It's not.

A

Anyways and just saying necessary and authoritative thing so.

B

To quickly remind everybody and.

C

There's obviously lots.

B

A

Different influences on the local field potential, and so we just go through them, explain a little bit. Geometry has to do with it and.

A

D

First, one is quite.

C

A

Nothing transmembrane currents. So this idea that when you have sign up seas, there will be currents so for.

D

Example, excitatory.

B

A

Open and so you have sodium and influx so cut, so you know for whenever.

B

There are synaptic inputs whether they make menorah.

A

A

You will deplete the extracellular medium.

B

A

Sodium ions, maybe also of calcium ions, and that means that whenever you excite, whenever you activate the synapse you're gonna have a negative local field potential outside of the neuron, which would be pretty obvious. The interesting thing, and that already goes to.

D

What's the the spatial nature.

A

Out is that the these fancies are not symmetrically placed around all of the neurons.

B

A

Are local so, for example, all these inputs.

B

A

That draw up the draw of sodium from the extracellular material right. They.

E

A

Far away from the rest of the cell, but the entire cell is electrically connected, while it's membrane, which means that this uptake is going to be local. But the field.

D

A

Right of this is spread throughout the neuron, meaning that for electro neutrality,.

D

A

Will get you will get sort of negativity here, but then the return current has to come from all over the membrane, which means you get a dipole of sorts. We will revisit that when we talk about the electrophysiology.

E

A

First point is just to know a big impact on on the activity.

F

A

We go to figure to a think. It helps with.

F

A

Most of these speakers also come actually from other papers they're, given that this review paper, I, guess and I found you know one of the other papers I wanted to show us actually where one of those figures came from, but we get to that. So it does go quickly. In this way, the.

D

Point that I made from above.

A

You see on the top left, there is.

E

No long and it's getting.

A

Excited somewhere near the apical dendrite- and this is a simulated model, but the good thing is.

E

Our simulations are very, very good for this, like.

A

The forward properties of the local field potential are ever perfectly understood or very near perfectly understood, is not really much of a mystery. So we can trust these simulation models and what you see there with the lines is, essentially you see the drop-off and the extracellular fluid near the excitation at that blue dot there, where we emulate in the air and the epical temperate.

A

No, that's that's. Yeah focus on the left.

F

A

You see those field lines which essentially show you where the amplitude flips from majority negative to majority positive. So you see that while the effect of the input is local in terms of making their exxor cellular material, so negative, depleting.

D

A

Of sodium ions right, you have of course, a return current over the entire norm, because it's internally connected so the amplitude closer to the soma is actually the other way around, which is not to be confused with the fact that this is still an excitatory current. So if you were to record the the intracellular membrane potential, of course, an excitation at the synapse will have a positive effect throughout the cell internally, but the field effect is different because you have an extra cellular, ionic flux.

A

So I hope that is reasonably clear. I think you can skip the the frequency part on the right. It's simply just the notion that that the frequency response of this is is there's a low-pass filter to this. So the high frequencies are only measurable.

F

Close to the dendrite okay.

F

A

Importantly, figure to see which is sort of the opposite of this. This is inhibition so and in this case there don't use stimulating current, but they simply make basket cells fire near a pyramidal neuron in hippocampus. And what you see is that because, as you probably know, these pasta Purcell's tend to provide para somatic inhibitions like very close to the soma and when you record the field potential across the depth of this neuron. So you can imagine a long long, needle with lots of electrodes along the step passing by the neuron.

A

Then you record different local field potentials depending on the depth, and what we see is that the extracellular medium goes positive and it's amplitude close to where the inhibition then actually occurs because of the basket cell innovating. That pyramidal cell near soma.

B

A

Then, when you go far enough out that amplitude switches so like nearly near the apical dendrites, the field effect is actually not positive, but negative. That is also shown by this blue line, which is normalized there on the central axis, zero.

E

A

So the staff that is close to the site, where you inhibit, sees a positive deflection of the field potential where it's far away. You see the return current because, of course, the neuron is internally connected and somehow a charge needs needs to come from somewhere. So that is just to show that these neurons form poles right, dipole, quadrupole, and so that the geometry is important, particularly when it comes to this effect of the synaptic membrane.

F

Just going back to the sodium spike, you mentioned earlier large effect, where the reason they say it doesn't show, because it's short yes like less than two milliseconds, so it's not like an action potential doesn't make a big effect. It's just short when you're averaging out these things doesn't work for a well I'm, just jumping and you another clue. This whole dipole thing doesn't seem to be very important in terms of usefulness, for our work I'm going to point out, they promote self will.

F

Have this tell a soulful not because it's this dipole effect here has to do more with the fact. We've got this out there, but.

D

F

Out that in different animals have different effects so in going from rats to the demise to rats the primates humans and so on, the dipole effect has more effect in the lesson.

F

When you line up a bunch of parameter ons, if they're all lined up on the same level, they will add, but if there are different levels, they kind of cancel that so the bottom of one is near the top of the other and so on. So you see very little of this dipole effect in primates.

F

A

The question is across: why would we care about the LP, anyways and one of the reasons, but we kept talking about the LP was because of its signature defect in the whole navigation system. So the fact that the whole hippocampal entorhinal cortex loop has light dominant sita and you know it that plays some role, and so the question was, you know: do we need to properly understand that and that's how we originally got into this hope.

F

Yeah so I love, you I, don't think it's that important. The.

A

Questioners doesn't do anything functionally and some people are arguing, for example, that effective coupling like field effects play a role in synchronization of oscillation someone. But there's no.

F

Mammals, you will see it might be really strong.

A

E

A

Maybe min we skip to the other paper.

F

Yes and the sentence is the temple, the corner, D coordinated, silent down, stood nearby, not just in for Grandma there's a negative in the super granulator, so they're basically saying there was an effect here of opposite polarity in the upper layers and the lower layers of cortex, and that was sounded like that was a persistent of fact. It's there all the time and sort of these up and down state. So it's almost like the dormant up and down states. That's what the after plant polarization word is a long word for something.

F

So this idea that the alternating positive and negative effects on the upper and lower layers- you know- maybe that's something that useful, maybe there's different cycling- that you know we're doing firing up here, don't find idea by firing up here in firing gun. Yet so the down states alternate being upper and lower layers.

F

A

Can we skip to the other paper.

A

If you scroll down a little bit, there is a figure like. Oh, this is already interesting just to quickly look at it. So the fact that that the theta reverses its polarity reverses its face, there's a full phase reversal along the depth right off this electrode. So this is just the rat ca1 probe. So you have all these different electrode sites along that depth, and you see that the field switches around.

F

A

Is pretty reliable? Finding, yes, and that has to do with with it geometry of this and there's models that have been proposed about how theta comes about? Can you scroll a little bit.

F

Pictures showing the actual probe like.

A

That is showing the actual. Yes, if.

F

That probe is not orthogonal.

F

E

Night, sweats q. Well, that's more! That's why these look.

F

At the between 10 and 14, or something like that is pretty strong that that's the little words I'm interpreting that because you might be seeing this shift govern artists on too late, and so, like you know, like moving across columns on the cortex or something like that right. So I guess. You already said that this is a reliable effect. Yeah.

A

F

Me feel confident in that. That alone would explain that, because that could be explained by a horizontal shooter right.

C

So if you were to average all of these guys out, you wouldn't see much of a theta rhythm.

A

Question no actually it's been shown that the so that goes to the models that are used to try to explain theta, which is a dual driver model. If we can scroll down a little bit and actually they're not perfectly in face, so you will not get a nice null incline.

A

So this was the little thing that got mentioned also that wasn't commented on YouTube so that these Broca's area, what so called the MSS MSTP, is the medial septal diagonal band night near near near Broca's area and where this so-called pacemaker neurons as I say that in quotation marks, because that's here is still highly contested. But the idea is that you can nicely explained this reversal of the polarity. If you assume that the pyramidal neurons at C, a1 or driven by two.

B

A

Oscillators one that drives inhibitory at the source so close to the soma and one that is at the down at the dendrites there. There is an excitatory source through the perfect path from intranet cortex and what you would get in that kind of a model, then is you would get an inversion of the face along the depth of this and you will get very nice reliable, theta synchronization.

A

They impart sort of refute this classical model of extracellular theta or like at least they point out all the problems with it, because there are problems with it, which is also by asset, and we got I thinking. What's your question on slack, wait.

E

D

A

Still debated and not entirely sure, on the whole concept. Well, Pacemakers makes me personally very uncomfortable because the brain is.

F

Pump up later, this is Jung proposed model you're questioning it other people question this start of the big picture. Are there now? Is there any agreed-upon explanation or is either mobilize but nations that are or.

A

Questions different different proposals: have you made lots of different mechanisms for generating theta exist and have.

F

Any of those cell then, like that's everyone, knows.

A

About I am aware of and I think one of the problems is that people want a unified.

D

Model of worse in.

A

Fact the brain is messy and have generates the del by different mechanisms in different areas, that that is my idea that my conceptualization yes.

F

A

I already, we already mentioned that many of these neurons in the cortex have these ihn ID card, which allow them to intrinsically oscillate.

F

A

We can scroll up to the next.

F

A

You don't need no actually, next one after that, I think four or five.

B

Although you show the.

A

Current sinks, but still further here we go right. So this.

E

Is an intro tendency to oscillation you.

A

Use a QX 314 to essentially prevent the neuron from firing, and then you step wise, like inject current to to change the holding potential of a dendrite. And what you see is that once you pass, you know next, 150 million volts minus.

D

50 millivolts or.

B

Something you see these very strong.

A

Theta oscillations, which become very dominant as the arrows that are marked sort of calcium spikes that are.

E

B

Then you come very.

A

Obvious in that lower figure B, where you don't block the neuron so.

F

Can we stop for a second? Yes, this was in the previous paper. We do not apologize and there was a couple of references in there that I thought I wanted, but I can get to. It was under these intrinsic currents and we're saying that it was only under certain conditions. Do you get this? What.

A

Exactly I wasn't chopped.

F

It was a paragraph like with intrinsic currents and resonances page.

F

Right, okay- and this is where it's talking about in the middle of that paragraph- is a sentence. It says for intracellular. Depolarization is sufficiently strong. mmm The rezident property of the membrane can give way to a self-sustained oscillation of the voltage in oscillations that theta frequency have been described exactly.

A

What you saw might know I know so.

F

The question I'm saying it was so it was in this previous paper, so the question was sort of curious about this night night I needed to read more about this and said it always had her somatic inhibitory, neurons haven't progressed into camera frequency, wasn't it it wasn't clear. These occur naturally or under these experimental conditions where they they said you have to have sufficient. You know that you are saying: there's some manipulating the neuron to get these things to do this, so I needed to be these papers, but it wasn't clear.

F

This is something that's observed, all the time and naturally-occurring neurons or it's like. Oh yeah, yeah,.

A

There's this one thing: one should point out where there's a difference between the idea that you get an intrinsic oscillation from this, and the fact that you just can't resonance, which has, which is the different phenomenon, initiate.

F

My challenge that resonance would be some you might die out. Is that right? How do we know that it means.

A

That you get a neuron that responds by an order of magnitude may be better when the inputs come up specific frequencies, so yeah.

F

So residence is not self-sustaining, oscillation! No, it's not a cookie to run well. That was the big question. I guess. The big question was that was a big distinction where, whether they do these things actually oscillate on their own or do they just have this resident, which is like a preferred frequency? So if you bang them they'll, you know they'll eventually die out, mean you keep going by keeping pumping right. So I that wasn't clear to me whether neurons actually create these oscillations self-sustaining oscillations.

F

Do they say it can give way to self sustained oscillations, but under certain conditions, where.

D

You're, just a.

F

Resident frequency of the agenda of a dendrite and therefore, if you.

B

Hit it at that frequency.

F

You don't get better responses, so no they answer that question whether these are self sustained oscillations. Well,.

A

I made these things are always complicated in vivo, but what is well-known, for example, is that when you change these ih currents, which is known that they are more strong, what is it along with also ventral or gradient of entry on cortex, where the grid scales and the theta changes? You can show that the these recorded threshold number oscillations change in frequency and they.

B

Change the frequency.

A

In proportion to the grid, which makes that a hypothesis for the.

F

More likely scenario is that there's a resonant frequency too.

F

And you had a bunch of cells that are fine connect to each other. They will more likely end up filing and firing at the resonant frequencies, because anyways a little often is that often a news but later was back end up doing that, and so the the fate of oscillation that would come about because of neurons would naturally try to fire together in that residence, as opposed to it, seems less likely. That would be a self-sustaining oscillation in the dendrite right. So I'm gonna go with that.

F

Unless you told me that you know otherwise, no I think it's.

A

Great I think it's fine, okay,.

F

There was talking about these self sustained oscillations and I. Think that would be I never heard of that before right. So that's what I mean.

A

I am the hospital to try enough aware of you know in vivo Pascal recordings from different dendritic segments to show that there are in fact, okay.

F

So, let's just go on now you go on that assumption that this is a resonant frequency and that these theta, then these beta oscillations, would arise, sort of naturally I any population neurons that are basically firing together, which I think it a really interesting idea because I've, never it never occurred to me before that's the way, without wait.

A

And the interesting thing is also that that is a can be a local thing like to the dendrite right. It's not I mean it can influence all say firing, but it doesn't need to well.

F

Yeah, but obviously because I'm going to see the state of frequency over across of our broad area, but if I'm doing an LLP and each dendrite segment, it's got its own little awesome little residents and they're, not in sync I'm, not gonna, see it shadow.

B

I think it's also what's.

F

Also easy because we talk about theta as a way of encoding by speeding up and slowing it down coding, motion right, so I think there's a coup yeah, but there's a clue there, because, as you point out, so you guys want to discuss. They actually want to oscillate together. If.

B

I sell part of the kid myself are.

F

A little bit early.

F

Because might be fun fighting, and so then what would cause you, so how do I get all these guys to fire faster or slower, basically, no to encode speed of interaction, whatever.

B

F

And that's an interesting question. You just pointed out you permanent building these things physically to change that way right, but I just.

A

It's been shown that like feet ourselves, that, when you deploy.

F

A rise in ourselves so.

A

Cells have have an intrinsic threshold, theta oscillations so called the subject.

F

A

E

A

Potential oscillation.

F

Meant T cells exist. They.

A

Call theta salesmen.

F

A

Was talking about like these dead mats are not talking about somatic effects so I'm talking about this, the specific class of cells that are called theta cells, which some people have been hunting for using different methods right.

B

F

Certain brain areas where.

A

We have found those kinds of cells.

F

A

Sound special this measure, a soft threshold oscillation in the center range, doesn't make the neuron rhythmically.

B

A

Of course, the interesting thing is that when you add current to these neurons that.

B

Oscillation can speed.

A

Up so so that would give you exactly that's why I mentioned the.

F

A

Thousand eleven cells- hey I, talked about that there well a 2011 paper which showed exactly these properties that they were looking for cells at how many? Yes, when I'm talking about my tea and I, show the. If you remember the vco law, I've talked about the fact and then there's that they specifically went out to find that speeds up.

A

Because these are the two hybrid.

A

D

F

You know I think in terms of.

D

C

I suppose, with the theta.

D

Oscillation in.

F

Intrinsic membrane sub-threshold oscillations in thumbs, you.

A

Can also look up well, they at out mm. hmm Oh that's very specifically. Oh.

C

That's the part, that's the living nuclear that project to the hippocampal areas. Oh, so it's like LGN projects to be one anterior, the anterior to project to this.

A

F

A

They listen Canada again.

C

E

Don't know exactly.

F

I would just look up the will. They help it.

A

A

E

That particular the ones in some of us actually obey the.

A

B

F

F

F

F

F

And the question is which of those cells any are how this property would should be sorted? You know.

F

A

Would be a really good questions? They don't further specify that in the whether people as.

F

A

F

The whole thing I think yeah.

E

If you want to understand that sucker, particularly if you want to use them as more general inputs.

F

No papi. No. We talked last time about the idea that perhaps cortex is unique in the fact that it has one of the makes it different than the older temporal regions is that the evolution I discovered that you needed a way of adjusting time and and if we need a centralized timing signal where you don't. That's Cal its timing scaling. But you may not need that in the end, because if you're mapping out environments, they generally don't have to do a scaling.

F

But if you, if you're, trying to see things from a distance or you're trying to hear melodies when or whatever you need that scaling, and so that may be one of the key weeks between out the house of all is to do this.

F

E

Those are exactly papers that.

A

Hunting down right now, as I mentioned, doing stand-up. So what's the specific question.

F

These cells that have intrinsic theta oscillations.

F

Membrane oscillations.

F

And again, I've always thought about the matrix cells as the timing signal, and so we.

F

Now get a piece of confirming these supporting evidence to confirm the supporting evidence for that apophysis for years and to know that that's how they work is very easy. Okay, it's something I did, but all the ways.

D

So about oscillations of theirs, pretty simple ways to create oscillations right.

A

D

A

Just a voltage dependent current yeah does.

D

It but I mean are these oscillations, are, we think, are we thinking about topology included in the methods.

F

D

The paper that I've been reading about emergent grid cells and necessarily talk too much about the topology of the structure. You know.

B

D

F

You created a topological simulation, which I mean it's really easy.

C

D

The natural topology to apply an oscillation through this space just from a random. You know, beginning place just by sort of doing some type of topological, meaning around to a decay function. A decay.

B

Or degradation over.

D

Timer of where do they call it where the firing rate just goes down automatically over time, and if you have that and then just using a topological mean around that point to reactivate it or to increase the firing rate dependent on that, then you get natural oscillation yeah.

E

A

Is a victim of ways to get oscillations and what.

E

A

To hone down on is you know what are the specific mechanisms with regard to grid cells and to what extent do those generalize to correlate with cells that are used outside navigation, so I'm.

D

Not necessarily.

A

Mediator, liner cortex as targets, but not necessarily you know, cochon with head Direction cells, but wider context. It was yeah because those targets of those cells are then putative core clear itself. You know so that you ghost was chasing after the smaller pollicis, which I would okay, so yeah I hope everybody's a little less confused a lot of peace. Now we're done yeah, that's good.