[00:00:05]
>> Ok, everyone welcome to the Second to last installment of the g.t.. No 7, our series her this and last year we're really excited to have John Runnels, joining US virtually from California civic time at the talk Institute. John has a leader in the field in studying those circuits and the mechanisms of visual perception, in particular visual attention.

[00:00:36]
He studies in economics as an undergraduate and then got his Ph d. at Boston University. And his post doctoral work was at the end I each with Bob doesn't own when he was still there. He really, that's where he pioneered a lot of informal studies looking at the circuit basis for soluble basis of visual attention and primates in the primary visual system.

[00:01:04]
And years after that, he moved to the Salk Institute where he's driving since 2000, I think 2001, Where he's continued. I Neer into really detailed circuits, and so he would be, this was for his Will perception visual attention. He was also technically that before front of what Recording in stimulating methods are used.

[00:01:29]
And primates in particular, I don't know to if you're going to talk about this today. But really I am hearing up to genetic methods and behaving primates. And also different types of neural recording techniques, and he's also moving to Use all of these behavioral cycle, physical and recording techniques and marmosets to really on a leash terror.

[00:01:53]
Hopeful genetic potential for studying this Hollywood race and some of the visuals and visual tensions. And John are very excited to have you here today and looking forward to your talk. Thank you. Thank you. And I want to thank you all for having me here. I wish I could be there in person with you all and each you maybe, you know the time.

[00:02:16]
I'm really very happy to be here though, and have this opportunity to talk about recent research in the laboratory. I'm going to focus much of the talk on a paper that was just published last month In which we Show that in a weight behaving visual cortex. There are travelling ways that regulate perceptual sensitivity.

[00:02:40]
But I'm going to begin my talk by just setting the stage describing a few studies prior to this in which we Looked at neuronal response variability. And discovered that The variability of neuronal responses is subject to cognitive state and particular that an attention is directed to a stimulus.

[00:03:03]
This reduces the amount of variability in and around the sick. And so, you know, the overall theme of the talk is this tension between this earlier work that showed that very belief is harmful reception is suppressed by attention. And that's Newark showing that a form of variability, Troutman ways can improve perception.

[00:03:28]
And how do we reconcile that in our finished with a theoretical model of that and bring these 2 So as of elastically, we were going to animal models of the 1st several results that are talked about are from the rhesus macaque. But we have to go more recently using our ancestors.

[00:03:50]
Well, in my laboratory, you know, they do have the potential to be developed ancestor net in animals. And there are a number of lineages that have been developed and will continue to be developed in the work that I'm going to speak about today. Really making you sit there, listen Baloch cortex which enables US to look at Entire areas of the brain.

[00:04:12]
In this case, area and t. simultaneously over the entire visual field to discern these, these tragedies sort of 1st. Here's a quick outline 1st, going to talk about all we're from Jude Mitchell and Christy Sandberg and might look at showing that attention, reduces individual neural response variability and follow up work in which they show that reduces low frequency.

[00:04:39]
Then we tend to a stimulus that reduces low frequency variation in neural activity. And then I'm going to touch on a more recent work from the Nevada 90 Who showed that if we use oxygen and it's to induce low frequency correlations, this actively impairs Sensory Perception. In particular, you to choose from a nation in that study.

[00:05:05]
Then I'll talk about this New we're Showing that fluctuations in neural activity in the form of travel in ways reduce perceptual threshold. And then I'll try to bring all of this together in a model which we call the sparse way. That shows that waste naturally occur under The asynchronous a regular regime, which if you're in sort of deep in the World, you'll know is an important idea in competition.

[00:05:31]
In neuroscience, and it's a state that has certain computational and edges and that the ways traversing the network that's in this state need not induce correlations or change individual neuronal responsibility. So this is we're from 2007, so it's very old now. But this is, we're from mental is now a professor at Rush.

[00:05:54]
Astor and Christie. Sunbird was a graduate student at the time in a lab. In this were we made use of a minute tensional test, in which The monkey was to fix a point in the Center of the screen. And then mentally track stimulates the moves through space. And at the end of the trial, the monkey with Group report which to me alive were highlighted at the beginning of the trial, so that some of the tasks that require the animal to attentively tracked stimulates the move through space.

[00:06:24]
And during their trajectory, the stimuli would cause at some point in time for about a 2nd, with one of the single eye sitting in a receptive field. So we could compare the responses of individual neurons in actory before when the monkey was either attending to attracting the stimulus as it moved through this space or attending to other stimuli.

[00:06:45]
And we saw that ship the typical form of my solution, which is an increase in being fire rate. So here you see more spikes on the monkey, attended to the stimulus in the receptive field, and when it is and it away. But we also notice that neuronal variability, it was lower in the market and into the perceptive, but this is particularly strongly modulating neuron.

[00:07:06]
There was actually a reduction in the standard deviation of the Spike out More generally, we saw that the fan a factor, the variance in the Spike out by the neurons response rate mean rate was reduced by attention. And here is an index, just a difference or some index of the fan of factors across the population of before US that Judy Christie reported with the neuron showing statistically significant changes in found a factor showing the black.

[00:07:39]
And so this was a very robust Kindy, and it was the 1st study showing that attention, or any state regulated variability of neural spots. So Jude and Christie then went on to do additional analyses with a reporter from Harrison neurons. And look at their variations that this is an example of what you might see.

[00:08:03]
This is actually, these are reportings in anesthetize, cack, the one you own. And each life here is the spiking activity and it's one neuron and where she can see it. There are periods of quiescence, separated by periods of higher activity. And so if you imagine reporting from a pair of these neurons, you would see correlations.

[00:08:24]
Now these are perfectly robust. These are show, this is NASA ties and all. But we do see correlations between pairs of neurons in an area before. And so we thought about this in the following way. This is a figure that's adapted from our back lay them and who she is review of correlated variability.

[00:08:45]
So I'd like you to imagine if you're recording from a pair of neurons, neuron one in 2 here. And we've presented a stimulus this, right? We're leaning grating to the 2 stimuli. I mean $2.00 to $2.00 visual system in reporting these 2 neurons. And each point, each black or dark Blue point here is a firing rate on an individual file.

[00:09:06]
And if we then present another stimulus, a left leaning, braiding the, the responses are here shown in Orange, and by, I'm sorry, construction, you know, these distributions are positively correlated. With one another and these neurons don't know both prefer this leftward leaning rating, so they share tuning in common with one another.

[00:09:27]
If we superimpose those points, you can see that because of the correlation between the response to this trial and trial, there is a degree of overlap. And this will happen in fact, of reducing the ability of an ideal observer. And presumably the audio visual system itself to discriminate between these 2 stimuli.

[00:09:49]
But if we struck all of the points so that they are now in deep correlated, then the degree relation is. This is one example, but it's just intended to illustrate the idea that that correlations among neurons that have shared preference and impair your ability to discriminate single And Jude.

[00:10:09]
And Christy went into the same database that they have recorded cell by cell. And they looked at parrots, and they found that in before there was a fairly robust agree of correlation might trial just like in the hypothetical example. And they found that when attention was directed away from the receptive field, that his hearer directing attention the means circle to the left while recording from a pair of neurons in the right visual field.

[00:10:39]
There was a higher degree of correlation. Then when the monkey was directed to its end into the They quantified this over many pairs of neurons. And this blood on the left shows the Spike Spike coherence of these pairs. And with data shown in Blue for the unintended condition of the attending dish.

[00:11:01]
And so what they found was that very robust reduction and a degree of Spike, Spike coherence, or which is a measure of where the nation Particularly at frequencies of 5 degrees $55.00 per or so. But to some extent it, it's all the way up to about 20 years.

[00:11:20]
And notably, we saw no change in coherence Ivory. When we looked at this in computer, just the mean value or than me, sorry. The change in The correlation in these low frequency side of the correlation was reduced by a factor of 2. So there was a substantial reduction in the amount of correlation.

[00:11:41]
And we and they acted analyses that Challenge is introduced back in the ninety's to quantified it do Re sort of signal to noise ratio as a function of the number of neurons that we pool. So this is a sort of a hypothetical construct. It's a model and so it's fraught with its own assumptions, but it enabled US to come up with an estimate of what the impact of the speaker relation was on.

[00:12:13]
On a point on the signal to noise ratio of the, of the, of the Neural signals. And before, in other words, how well couldn't I feel observers from any stimuli from one another. And what we found was that when we Presented it, when we, when we stripped out the changing correlation artifact artificially, and we allowed mean firing rate to increase as it did.

[00:12:39]
This increase the signal to noise ratio by about 10 percent at asymptote. A large number of pairs, But when we Instead Maintain the mean firing rate that allowed the decorrelation to occur. This increased the signal to noise ratio by 40 percent, 39 percent. So the kind of closure of this is that about it, about 80 percent of the benefit of attention is attributable to this reduction in low frequency correlated variability.

[00:13:12]
Which was something that until then you know what it really even looked at. Although I should say that Marlon Cohen and John Mansell were looking at the same problem. And very soon after we published our work, they've published beautiful work that's very much in agreement. So on another, you joined the lab at that time.

[00:13:32]
And we had just developed some of the techniques that allow mentioned in which we Didn't enable US to do up a genetic some primate effectively. And so we wanted to test this empirically. It was a great deal of theoretical work that had been saying why correlations could be harmful to perception, but no one had ever tested it by inducing correlations.

[00:13:53]
And so this is a we're of on or on Monday. And Joe's are Massey underpants. Now, professor at Yale and on and on. And Joe, Joe is a senior scientist in scope. It's The 2 of them work together on a project in which we thought we could introduce correlations by lightly tickling the cortex with a laser that fluctuated even at low frequencies in the range where attention actually suppresses correlations or at higher frequencies.

[00:14:20]
We're at no effect. And so the idea just conceptually, as it started, if it's tension, has to be correlated the signal when the monkey needs to attend the stimulus. But we reintroduce the correlations that are removed by attention. And we show an impairment and discrimination consistent with these sort of theoretical models.

[00:14:40]
So the approach that we took in this work was to remove the dura mater that protects the cortex over area before and replace it with Silicon material. This is based on an approach that and a rope developed for Or intrinsic imaging. And it's worked beautifully for US. In fact, for those who are interested, I'd be happy to help you get started using this it.

[00:15:07]
So we don't have any monkeys that have intact or is it more Because it removes the need to remove Jisshu that grows on the surface of the door of which can be or the Bane of your existence if you're a primate neurobiologists. But what we can do here is we can interact virus through the Peel surface And then cause an expression of our sins.

[00:15:31]
In this case that we are using our sense that we are driven by the campaign is promoter. And so they lead to selective expression. And science for itself, and then after allowing time for the virus to, for the protein to express. We could then go back to the site of the injection with a fiber optic cable connected to a laser.

[00:15:50]
And we could regulate the neural activity. And so the paradigm was, this follows, the monkey was presented with an intentional cue directing it to attend to one location or another. And then within the receptive field, we present it as a target. And there was a similar target percentage at the other extra receptive feel confortable receptive feel.

[00:16:12]
And over several presentations, the monkey would see an orientation change and needed to this district discriminate that change The manipulation then it was to induce correlations at either 5 to 40 person low or high frequencies were attention. It's not suppress correlations and ask if this led to a change in the, in the monkeys ability to perform the tasks that we perform, the exact same trial randomly entered into taking low frequency and high frequency laser stimulation.

[00:16:50]
And we could with out, strongly regulating the mean firing rates of the cells by just let me take a limb cortex with very modest fluctuations in laser intensity. We can induce correlations between pairs of cells. So this is these plots to individual cells with base locking. That's fighting, that is lost to the, to the, to the fluctuation of the laser intensity.

[00:17:19]
And I could show you that this did introduce correlations between cells. And the main finding was that low frequency stimulation doesn't care a performance. So here are 2 examples of sessions, in which low frequency laser stimulation you carry the ability of the monkey to, to do the orientation discrimination task.

[00:17:39]
So in grey, we show the psychometric functions as a function of the orientation change magnitude. And you know, you can see here that the monkeys were at 75 percent or so with modest 2 or 4 degree when you, Titian changes and then superimposed in Blue, or the psychometric functions fitted data recorded when we were in using these low frequency correlations.

[00:18:06]
And we, and we see that across the population of recordings, there is a, there is a Measurable and consistent reduction in my attention discrimination. Now this is a fact was spatially specific. So here is a 3rd section session c. showing the same pattern. And rightward shift of the psychometric function.

[00:18:28]
But on the same thing in that same session, when we directed the monkey to attend away from the site or reducing this relation, we saw no measurable change in you, nor your patient discrimination at this other sites are specific to the site where we injected the virus and were driving before elections and it was also frequency specific.

[00:18:49]
So we went to the high frequencies where At 40 Hertz, where attention does not relate correlated variability in our earlier work. We see that there is no effect. And these and similarly, if you expect in the markets as a way that's not a fact and variables. So this I don't have time to go into all the data.

[00:19:16]
But the correlations were comparable magnitude. So we titrate the laser, so that the magnitude of the correlations that reduced that these 2 proven seas were very similar to one another. And from this we can conclude, I think that that low frequency correlation does in fact impair perception. And so we were very happy to see this result And it fit nicely into a tight little bundle with our earlier findings.

[00:19:42]
And so we, you know, at this point we felt confident that we could say that, that the reason that we see attention to any production seen carrots is that these low frequency correlations are harmful to our session. And that's all fine and good, but Then we stepped outside of looking at pairs of neurons and we started to look at entire populations of neurons.

[00:20:05]
And so this is more recent where it was just published Last month. And this is the work of a really wonderful post Aat Zack Davis, who is by the way, going on the job market soon. And another person who at the time was the most talking Terry Sejnowski is laboratory.

[00:20:26]
And this is a collaboration between Terry's lab, my lab, as well as with important health from Julio Martinez through the And so the idea here was to report using a Met a method that enables US to report process a population of neurons. And to look at the space of temporal patterns that occur.

[00:20:47]
And so we implanted huge heart races over area and see in the common marmoset and, you know, the marmoset is virtually listen about. There is one Big Senate little wrinkle, but basically the whole horrible service is sitting right there into the skull and is available for recording. And so this is the 1st time in it we that in and awake animal that anyone has been able to look across the entire visual field of empty Consul adults of your right and see what's, what's going on.

[00:21:21]
So this is Part of every report of or were single units and multi-million activity. And this is an example I think of multi-unit recording from a single electrode that was aligned with the position of visual stimulus. And so you can see across many, many trials, there's a period of, on the left of the line before the stimulus appears, where they're sort of very well spiking at a low rate.

[00:21:49]
And then the stimulus appears at the time of the vertical line. And after a brief delay, there is a robust visual response. And that's one measure that we, that we can take from the, from the top array. But in addition to that, we can record low field potentials which are just lower frequency taken from the home, from the full field attentional.

[00:22:14]
So we filter these from a one to 200 Hertz, and below feel the potential is a measure that is somewhat, but it is a measure of the, of the change in voltage in the vicinity of the electrode. And it results from the return currents that occur as neurons are active.

[00:22:37]
So there are sent out to currents and membrane Currents do just fighting that Sun. And so it's a measure that enables you to look at small populations of neurons around the tip of the electrode. And in fact there's, we're dating back to the seventy's in which simultaneous recordings, using intracellular approaches have been combined with low field potential recordings.

[00:23:05]
And if you conditional, I suppose, field recordings, intracellular and low field recordings on the spiking activity, you can see that their marriages of one another. And so I think that a reasonable way of thinking about a lot of feel the tension is that that it is something like a measure of the intracellular potentials across a population of neurons in your head.

[00:23:27]
But Elector. Now for those of you who record in these, these little feel the tangible Who have recorded a lot of the potentials, There's a look familiar period prior to the setlist. The average flow field potential is very flat. And then at the time of the stimulus vote response, or all of the neurons or population are answers, finding the stimulus, there's a Big deflection That you can easily see.

[00:23:54]
But in fact, we look at individual low fuel potential trades recorded on a, on an individual trial. You're shown pink. You can see that for this neuron, you really, if you were asked when did the stainless occur, all you were armed with was that was the paint trace that local field potential from that individual trace.

[00:24:13]
You would be at a loss. And we can look at this in a couple of other examples. So this is a, another, another. And he, you know, on every recorder in the marmosets, it sat there looking at the screen and waited for a stimulus to appear. And you can see again that this, it was which was 2 percent less than 2 percent contrast.

[00:24:34]
It was just at the perceptual threshold of the monkey. You couldn't see it in a mean. And you couldn't really tell. And it occurred in the, in the individual trial. And this is shown on the right to speak of plots where we're showing the relative power of the local feel, the tension prior to, and after the appearance of stimulus.

[00:24:56]
Each model race is for a single stimulus presentation. And the Gray Point that's a little larger at 0 is the, is the mean of that population. And so you can see that individual trial, the mean is sitting there that the mean Of the power across all of these is right in the middle of the individual trial.

[00:25:24]
So you really couldn't see anything. And the black is the trial British response. So even in the accurate, if you look at the relative hour prior to and after the appearance of the stimulus in the mean response and local feel potential, it's still sitting right in the middle of all of the problems.

[00:25:40]
And so you kind of make it out, but it's very difficult, even if we increase the stimulus to a very visible 10 percent. You can see that although in the mean now the black point in beehive and black trace, you can make out the stimulus focus on The The Individual points of the being higher, although shifted upward now are, are still quite variable and it's still difficult to see.

[00:26:06]
And so the main point that I'm making here is that if you look at the potential of an individual trial, the variability in neuronal response, which is a measure of the variability of small population of neurons or under electrode, is quite variable. And it's on the order of magnitude of this thing.

[00:26:24]
Let's look at science itself. You cited to that in the, in the movie for Mad cow, and there's a lot of fluctuations. But what we're showing here is that even in the Awake animals, there are fluctuations in the population. And so the question is, what do we make of those large fluctuations?

[00:26:39]
What is there? And here we apply a technique that was developed originally by a lot more work and colleagues prior to this work. And it is a statistically rigorous way of just detecting spatial temporal patterns in ongoing fluctuations like this. And so the, I won't go into too much detail, but it's a briefly summarize.

[00:27:06]
The idea is to record local field potentials across the entire race of each actor as it's a measurement. And then you compute the gradient of the phase of the low feel, the tension across space. And then you look for a points that maximize the divergence of that. In other words, points that are like shown down below on the left points where there is A great change, a massive change.

[00:27:35]
And in the, In the gradient of the look, feel potential feel. And these are then make up candidate starting points for travel delays. And I'll tell you then what we do with those. So the 1st problem that we encounter though, is that these fluctuations in love who feel the tension, are very prospect don't occur within a given frequency range.

[00:28:04]
And in fact, the frequencies can change very quickly over time. And so the polar transform that one would normally apply to A signal to determine its phase does not really apply in bright spectral signals. There are problems that are, there are violations of the assumptions of the over transform.

[00:28:23]
And so while the Delta method, which are a measure which we refer to as generalized phase and then gets around the set we've really little bit beyond the scope of this Talk to go into detail. But essentially what we're computing here is the phase That Within the dominant power and any point in time.

[00:28:46]
And so does that bias anything? Well, it really does. So if we now look at the degree to which spiking activity is Phase locked to the low field potential, we can compute that phase locking and ask, you know, how related is the spiking activity to the phase using different measures of face.

[00:29:07]
So we can Compute this end in narrow frequency ranges, which is the more traditional approach because of the problems that I mentioned with the Hilbert transform. And the fact that if you pick a frequency range, you're going to Miss the power outside that frequency range. In this signal that changes its root when c. over time, you can see that the Spike face covering, or for particular bands of local field band has filtered local fields, are much lower.

[00:29:42]
There are less predictive of the spiking activity than we have with this generalized station measure. And this is partly due to the fact that the frequencies vary over time. But even if we look within the small number of trials, but there was significant power within a particular band, which is only 12 and 7 percent trials with these 2 bands.

[00:30:05]
The generally space measure applied to every trial still provides a better measure of, of the relationship between spiking activity in the local field. Actual dates, though, we believe this is a New measure that people should pay attention to. And it's described in the paper that we just published and happy to talk to people about how that could be incorporated work if you're interested.

[00:30:26]
But the point is we now have a robust, meaningful way of thinking about face or Rod spectral signals. And so we can then go across the array, compute the, the generalized phase of the local field potential at each, each electrode on the Ray. And that the approach then, the statistical approach is we compute the change in phase as a function of distance from the putative source.

[00:30:55]
So this putative source being the maximum divergence of the gradient. And here is an example in the middle lower at all of cases where there was a progression of phase away from this source. We compare that to a shuffling of the data to create an all distribution. And if the, if the changing face over time over, I'm sorry, distance from this, from the Puter to source is well outside this knowledge distribution.

[00:31:28]
We then assert that we found a way. So Here is an example of what it looks like in a simulation. So this is just noise, the Red dots are candidate, we have sources, but when we actually appears the dark will turn Green, you'll see a Big wave go out and these are made to be obvious, reduce C.

[00:31:49]
Really the problem is that in the awake state, the data is very noisy and the size of these waves are such that it would be difficult to detect. It looks nothing like this. But this is really just to illustrate that false positives are rare. We can detect actual wave patterns.

[00:32:09]
What we see in fact looks more like this. So these are unfiltered. Not spatially filtered, Actually, it's a recordings across the entire it. So each of the electrodes Here is an individual local field potential voltage. And these pole changes have been selected from times when we detect it away.

[00:32:31]
So this is the sort of thing that we see. You can see here a way that appears to be moving somewhat downward to the right. And the Color coding from Blue down to Red is a measure of the amplitude of the millivolts, of recorded on the, on the individual electrodes.

[00:32:54]
And so this is not spatially filter, you're not seeing something that is artificial thrust. We've superimposed the Gaussian on random noise or something. These are actually the Independent or the individual records local field changes recordings on the channels. And so, you know, the ways are not simple or not like ways that look like they're crashing on the Ocean shore.

[00:33:17]
They're somewhat variable in shape and they, and, and Their head of genius. They don't seem to follow any particular project or even from the periphery or anything like that. But they have a Big impact on perception And so 1st before we get to perception, these, these are, this is a plot to showing the probability, the relative spiking probability As a function of the generalized spacing of the local feel, the tension On trials where there were raids on individual rape trials and the Blue bars show the probability of spiking is a function of the generalized space of Love Field attentional.

[00:34:02]
And so you can see that at plus or minus high, which we think corresponds to with the polarized stay. There is a higher likelihood of spontaneous spiking activity occurring then at phase Euro, which we think corresponds to Hyper of all right state. And so there's this robust phase dependence of spiking activity.

[00:34:25]
Just as the monkey sits and fixates a spot on a blank screen. The Red dots show up control, which was intended to look at whether we were seeing also changes the end synchronous changes in activity across your reading. So it would be here are just average the voltage together across all of the electorate's computer there if it is.

[00:34:47]
And asked how the Spike in activity depended upon that aggregate average response load feel the tension value. And it was really not much. So we conclude that probably ways that we can detect are strongly correlated with spiking activity. And there is really no evidence or synchrony across the entire array Simultaneously occurring everywhere.

[00:35:17]
So this led US to think about, well, about of these ways, due to perception. 1st hypothesis is that sensory responses That occur during the polarized space of the wave, maybe potentiate though, here, and we're showing a picture of, of, and see, imagine that we have lowered the slack, trooped out and see this sort of hold and see recording of incoming spiking activity from your I'll record a Spike in activity at that neuron, and we're going to assume that there are some incoming sites That we'd better better opinion on when you're on that reporting from it.

[00:35:54]
Because at this moment, the traveling wave is such that the cell, it's in a Hyper polarized location and the map We have prophesies that this will lead to less Lightning activity. But as the wave traverses and C, and the neuron that we're reporting from Is within and we're polarized location and in the right topic.

[00:36:19]
Now the firing range can vote by the same. Incoming Spike train is elevated and final and you know, fully polarized period. Our hypothesis then is that the spiking output of this neuron will be, will be stronger potentially. Our 2nd officer says that the potentially Ation of the sensory boat response across the population, neurons there Increases the probability that a monkey will attack its heart.

[00:36:47]
So to test this, we have the monkeys fix a fixation point and perform a simple attraction task in which the target appear random position a random point in time, one of 2 potential positions and States. I was a little drifting towards stimulus. And the monkeys task was simply to make this a Caterpillar, the creation of target, since the monkey solved arc as a simple detection task.

[00:37:12]
And beauty would be something. And if the monkey responded within a short window of time following occurrence of the target, we reported the animal was trees. And so sack and lie and measure Monkeys, psychophysical performance. In here you could see 2 monkeys shiny and point straight and Red and Blue.

[00:37:33]
But the 2 of them and so they were able to determine for each monkey the level of limb in its contrast, that was necessary for the animal to attack the target 50 percent of the time. And. And we focused on that luminous contrast for the rest of the experiment.

[00:37:54]
Now, one critical thing that we worried about was that we do see a pretty large difference in firing rate. And in a point where the monkey to attack it's harder to not detected. So here, mistrials failures to detect the, the spiking activity was substantially lower if and when the monkeys detect that part.

[00:38:18]
Why is this an issue? Well, we know from earlier work that while did with Predator show and colleagues that simply presenting a stimulus within the visual field can trigger the propagation of the stimulus vote in the local field tension. And so we could really have pulled ourselves into thinking that these funds heinously generated by Aids, improved perception.

[00:38:44]
If we simply looked at the time of the stimulus, the vote response and asked whether the wave, because the higher Spike rates on trials would presumably lead to larger and more robust ways. And we could then escape positive fact. We could think that and heinously generated weights were causing a monkey to attack the target.

[00:39:09]
When in fact, they detected target provoked a stronger response and triggered a large wave. And so to remove that content, we have to do some, some, some careful analysis. And I will step through this, but the bottom line is, we said, well, Let's step backwards in time before the appearances chart and ask whether the waves are in a particular state prior to the appearance of the tarp.

[00:39:38]
And if you imagine sort of metaphorically waves propagating and you know your past, The idea is that, that if they are in a particular state prior to the appearance of the target, then they might be more likely to be in a particular state at the time of the chart, but the prior to the prior experience of the target, there's no way that the that the future singles over spots can reach backwards in time to fool US into thinking that we're seeing something systematic.

[00:40:08]
And so, so if the monkey perform this task, we computed and direct your attention to and be here. We record of local feel potentials on individual trials. And we went backwards in time and asked if there was a point in time prior to the appearance of the target where hits were hits, where the Taschen of the target was preceded by a particular based on line.

[00:40:39]
And so Indeed, we show individual field potential traces, and you can see visually that on the traces at a point in time prior to the appearance of the target, the, the local field detention tended to be at average state. So the phases were alight at that point inside and we can compute at that point in time what you see and see which is the crash of trials with a particular generalized face.

[00:41:08]
Other tended to be a particular face at that point in time. The parents of the target, and we found that this was true for hits, but not for misses. So if you look down below and the greater it says, the Missus also shows like 2 Asians and one field attentional voltage.

[00:41:23]
But they didn't show any preposterous of a particular case. And so we could compute this measure of Phase alignment, which is a sort of measure similar to the circular variance of this generalized phase life in c.. And we can compute that over time. And by comparing to and I'll just give you should ask whether there was a period of time in which there was a particular tendency of this, of the brain to be in a particular phase state for Hits and Misses and impact.

[00:41:53]
And you were showing these analyses for 2 monkeys for both markets. And you can see that the timing was somewhat different because monkeys showed a period of time when there was a highly significant face alignment prior to the appearance. That's part of the target. But there's only occurred on a hit trials and not in his trials And in the analysis and shown, and in each we compare this to a shot.

[00:42:18]
So we can look at the spent fuel both ways and ask, Well, was he above response in the low field potential, stronger on trials when there was this pre target phase alignment? And what we're showing here is it is, it is, is in Blue, the face alignment for the individual trials added together.

[00:42:40]
And then in black is a shuffle control. So the black Distribution also shows a preference for a particular face, but that is a reflection of, Of what you would expect on average from the, from the, from the Yes. And the difference between the Blue trace and black trace is the amount to which that's handed upon the face in the prepays and preach heart face.

[00:43:10]
And so I realize that that is a very complicated thing for me to explain in a brief thought. But the point is that prior experience at the time that waves are, have a tendency to be a pretty diverse state, which is revealed by this pretty target face alive.

[00:43:28]
And that predicts the likelihood that the monkey will, will, will detect the target at the time of the target appears. And so from this we can conclude that Through this indirect method, we can conclude that the phase of the low field potential predicts the ability of the monkey to subsequently detect a target with our, of the potential confound the stimulus vote.

[00:43:57]
Wave Infiltrating our measure of phase of life. So then if you get back to our 2 hypotheses 1st, the question of whether sensory responses oppose during the polar ice phase of the way for potential. We can see that after carefully disentangling the simplest approach component from the way my family did it, we can see that, that on the way to trials, the stimulus of a response was stronger.

[00:44:29]
Then, then, then it was in the, in the hypothetical space. And if you look at this across recordings, we can see that this is particular to wave trials. So the wave trials phases prior to the appearance of the stimulus are strongly predictive of the stimulus of motor spots.

[00:44:53]
When the target eventually curves, this is not to say that we use our own are the only thing that regulate the strength of stimulus, but response. But what it does say is that there are predictive so that we look backwards in time because there are spatial temporal, we organize, we can predict the future state of the cortex, and we can predict Of, you know, face that any way to spread stimulus of response so to put it very shortly, group briefly, if you regulate the strength of the sense of response and the stimulus of a sponsor, stronger during the depolarize phase of the way than it is during a higher price for it.

[00:45:32]
So then what about the 2nd part that says that this potentiation of sensory or both responses increases the probability of detection. Here we see evidence for this. So using The same math is that I just described with the physiology. We could, we could ask, how likely is the monkey to have a target as a function of this space?

[00:45:55]
And you can see that at the 0 or Putatively Polaroids state, the monkey was more likely to attack a target. So your column at this level contrasts the monkey detected the target 50 percent of the time. If the face was aligned so that the pole or its base was Present at the time a single Sunset.

[00:46:21]
The monkey detected target on the 70 percent of the time. Whereas in the higher polarized States, the monkey was impaired, perception detected more like 30 Five's So a very strong regulation of the size of the Monkey's ability to Act work. This is just the same data for proposed monkeys and comparing we Trust to just all trials, averaged together, which is what you would see if you didn't Differentiate between ways and just fluctuations in the local field potential in general.

[00:46:52]
So if you were reporting from a single actor and looked at this base measure, you would find it faces pendants. But we think that is because you are mixing in trials where there were and were not ways by looking at ways. We can much more strongly predict the Marquis of success attacked Mark So I only have a little bit more time.

[00:47:12]
So I'm going to go through the. I don't part little bit fast. And if you want to leave a little time for questions, but this, this really raises a problem for US, which is that we had a series of experiments that had built up to the point that we were convinced that fluctuations neural activity were counterproductive.

[00:47:31]
And yet here we have a form of neural responsible actuation problem weights, that demonstrably regulate sensitivity and give a brief moments of improved sensitive so they can have the fact that increasing your sensitivity to simulate it live in time. So they're helping you with your perceptual tasks. And the way of sort of cutting this, we're not, is to think about this in terms of, of, of sparsely and so briefly And relisten running a little tight here start that we began with a large scale of a 1000000 in her own network of neurons that were that had conduct in Space Center.

[00:48:18]
So a lot of computational power required, a lot of supercomputer time to do this. And we began with, with Publish work or read to pitch a little publisher in which we have a randomly connected network. Most people in the theoretical work don't really pay attention to the time that it takes action engines to traverse their axons.

[00:48:39]
And so a lot of them are, has been done, really treats the certified state as being our end of the connected network. And we could find a regime that generated what is called the synchronous that regular regime. It's high conductance States and neurons are potentiate and they're ready to ready to of responses quickly when stimuli her.

[00:49:01]
And you can see that and they have the ability to generate ongoing spontaneous activity so they can keep themselves active at an external input. And so what you see here in the middle panel are the Spike raster, across a sudden sampling of this 1000000 in Europe. And we're superimposing this mean rate Cross the network with the mean local field potential.

[00:49:26]
And you can see that there's no particular structure. And if we look at the low, any model of field potential, which is just something excited to hit or inductances with the region around each point and feel the tension measurement. You can see that there's really no structure. It was no spatial structure, as you would expect from a completely randomly connected match.

[00:49:46]
But if we simply impose tough policy to the Never so that neurons, neuron and other tend to be connected with one other used connection probabilities that are modeled on And happy. Then you see, you can maybe discern a little bit that there are cultural Asians that are, that are occurring with little increases and decreases in slightly probability of Cross the raster.

[00:50:14]
And you can see that there are fluctuations now, not in the mean rate, that there are also fluctuations in low field potential. But importantly, simply imposing this typology and the requirement that actually potentials take realistic times to propagate a longer acceptance. We can see that it can generate traffic lights.

[00:50:36]
So this is these, this is a times a frozen image showing you the voltage distribution at a particular point in time. These are moving time And because the network is large, we could produce a network which was our sleep connected. And as a result of that, we could find a regime and very large parameter regime and wish the difference between these 2 networks, the one it generates waves and the one that just had no impact on the distribution of fire rate shown in the lower left hand panel, or the prohibition of variation shown in the middle or coherence between pairs of neurons in the network by exposure parents.

[00:51:24]
And so this is pretty, this is sort of a beautiful result that says that you can impose the sorts of patterns that we observe empirically by simply incorporating distant dependent connectivity and realistic action, intentional propagation time. And you can do so in a sense, if the network is large enough in a sparse way that doesn't disrupt the spiking statistics or Paradise like in statistics of neurons in a network.

[00:51:53]
And so hypothetically, it's great, we could, we could, we could have ways we could have the benefit of an increase in sensitivity to polarize space with across the cells without imposing what we know to be Counterproductive. Correlations and variability in spots. So what did we actually do? To get to that point where we simulated 2500 different Simulations of this network with different values of inductance, but yet side, it's raining, it's very soon ****.

[00:52:27]
And we found a regime which is shown in this black outline, that in which the networks maintain a directive and average pen You know that you're giving a talk on a Problem. So we value regime in this, in this space of excited or inhibitory conductance, where the neurotic with a neural network maintains its activity in darkness without needing input.

[00:52:59]
And it corresponded almost exactly to the regime its rise to the asynchronous irregular regime. And so it's a very natural consequence of the connected network that have the ability to produce waves in the same parameters, again, generates this idealized asynchronous or very rich. Now you can see in the lower left hand out that there are, There are concerts for different size networks and this is really where building a 1000000 neurons after it's critical.

[00:53:31]
Because if you get down to a smaller network of particular The point 5 millimeter, you can see that the Blue trace, which just traces up, the lower part of the triangle is very and it doesn't generate. This was some most here. It throughout the months and smaller scale and where they have generated track in the ways that generated what we now understand are enslaves.

[00:53:53]
They need to have strong connectivity between individual neurons in order to generate weights. They need strong conductance about 10 times the measure inductance and houses. And under those conditions, we see a very different path. But before white get there, just a couple of reality check. So what does the network look like as compared to the data, put it in and so this our slate network exhibits qualitatively similar.

[00:54:22]
Distributions of mean Spike rates is what we see in the marmoset shiny Red in the upper right. The coefficient variation is qualitatively similar and in fact, neither of these 2 measures was physically sick, 50 different from the facts of the model, the distribution of the coefficients variation and they are Diller between them are set in the model And the wavelength distribution of the way says also some or so here in yellow we show that we have a distribution.

[00:54:57]
If we shuffle them all, feel the tensions that naturally gets very choppy and small. But if we apply our local field potential model to the, to the spiking activity, then that distribution, the cumulative distribution function of the wavelength of the ways I see along the data. And finally, the propagation speed though broader in its tail, in the data, is similar to what we observed in, in the markets at.

[00:55:28]
And so I thought, you know, I think it would have been impossible to imagine producing a model that was statistically insignificant to significantly not different from the data is a model is a simplification. But I think I'm very happy that the data and model are qualitatively in closer remit under the conditions that your price to to be synchronous original.

[00:55:49]
And if we go back to the smaller of all, in order to generate ways, we have to Increase the number of Sachs's between a smaller number of neurons and increase the undock Gensis to sort of unrealistic levels. But it was possible to generate ways, but they were dense waves that involved most of the neurons operating together.

[00:56:15]
And In fact, the phase Just review SION, of spontaneous spiking activity, shown in the bottom right well, was very different from what we observed in a period which were shown in the lower right, lower left. And importantly, the 2 must differ in the way that the travelling weighs a fact ID.

[00:56:40]
The stimulus of both responds. So this is my final slides, I'm sorry to say in a few extra minutes here. But here we are showing the stimulus of a response in the model for the sparsely Napper on the left and the dense, but a network of smaller scale network on the right.

[00:57:00]
And the black line shows the, the stimulus of a response to a small African input to the network. When there are no AIDS, The Blue light show the response was vote if the stimulus occurred at the moment. This wave was emitted polarized state at the location in the map where the stimulus was a vote.

[00:57:27]
And you can see there's a, there's a, there's a strong increase in the, in the Symbols of a response relative to the Hyper polarized state. That if we go over and out of the dense with network, you can see that there is a Fluctuation. It's imposed by the dance, but it's propagating across the never shiny black.

[00:57:52]
And that they really drive the network independent of whether or not the stimulus is present. And so the thing that we would've expected that you know, you can do things from for elevations is harmful, is occurring here in Martin density or, but in the sparsely network, we don't have that problem in attention.

[00:58:11]
A response of vote by stimulus. If the stimulus passed and it's either responding to the Polar Ice, The relative to the had before US and such, just really quickly to remind you of just the true, rather hurried way. We found that spontaneous activity is, is strongly regulated by the faces of trout and weights as it is the stimulus of both response.

[00:58:41]
And there is also at the same people, right space and increase of the monkeys ability to detect the target and theoretical model. A large scale sparse wave model can fit qualitatively the patterns that we use that we see and change and has the property that it can increase the stimulus of response if it come inside to the right space.

[00:59:08]
But only a large scale model that sparse connectivity has this property, and it doesn't, it doesn't disrupt the spiking statistics of individual neurons, work terabytes statistics measurably. And without starting treatment questions. Of course, great cost pose a really excellent talk trunk. You know, There were no questions in a chat.

[00:59:34]
Very funny one time, you know, you won't take a couple of extra minutes here for sure. Presenter questions? Yeah, I think they're not all out with them over Lucido. And this is, you know, what your audience is a complicated talking as of the Need to avoid that confound with the strength of the stimulus vote response.

[00:59:57]
And the whole little saying we have to take it to look backwards in time. And it's predicted this. And I think that's the only way you can really do it. So I'm afraid we're just stuck with that that. Let me go. Did anyone want to? And muted ask a question.

[01:00:16]
There's a, there's a question in China as well, but some lost time. I am a player and this is then about what so well, all of their behavior in a recording session which are beautiful, habited with relatively simple facts to mine. What do you think would happen and how do you think these corporate waves would interact with them?

[01:00:41]
Are natural complex stimuli that thank you for that question. Yeah, that was a worry that we had because there was work from the Taylor Kennedy colleagues, really didn't look at try file if they didn't have that technique to do that. But they computed site and look and feel potential across the entire race and in several animals.

[01:01:03]
And they found that there was a pattern that looked like it could be a troubling way. But they found that the amplitude and spread of the wait was very strong reduced by the contrast of the background. And so that were let US to worry about whether these braces are special to implant computer screens and therefore totally records in our perception.

[01:01:23]
And so we also recorded the same data as the monkey sat in front of a computer screen looking at natural images. And they were free to look around and see CAD. And we found that during periods of m and there were huge waves that were triggered by the change in the retina much but during fixation of naturalistic images, the waves persist and they're very similar in their statistics to the way that we see in a way that we are so, so I think these are basically an intrinsic property of the network that you can't get rid of by presenting a high contrast background or naturalistic scene.

[01:02:00]
And so they're probably relevance not such. I have a question. Yes, John, nice talk. I really appreciate it. I'm curious if, if the ways you're describing are they are the Dependence on across different animals or subtext. And I'm wondering if there's an effect of training or experience or anything like that that might just suggest other than just you know, what a phenomenon of cortical tissue Yeah.

[01:02:37]
So I, we're really interested in knowing whether these waves are just kind of the Ping that actually propagates Frost surface of that or tax it moment. You know, you know, you can imagine that it's just useful to have something that randomly puts you in the poorest state for a moment, but does it in a spatially way so that a population of neurons are momentarily potentiate and unable to see something.

[01:03:02]
That's our default assumption and the model that's really predicated on that idea that we're looking at some network property that's not governed by cognitive state. But we're very interested in that question. And in these animals, We didn't have the ability to answer that question. The statistics are pretty similar process to animals.

[01:03:20]
They had similar training, and it's Haskell's one in which to target that if you're at one of 2 locations equally probably. And so they didn't vary anything that would enable the animal if they have the ability to do this, to regulate weight, so as to benefit them in the context of the task.

[01:03:35]
All they did was they know, Hey, at some point something's going to appear and I better be my eyes. And so they benefit in a way that was not Was not subject to their cognitive control, except maybe that they were always sort of mentally monitoring those 2 positions in space.

[01:03:50]
But, but I think that's a really great question and we do have a beginning in collaboration with a couple of different Labs who are accorded similar data in the Cascade and discrimination. And the animals do have more information about what they're doing there. It may be possible to tease that out from those data, but I think the ideal kind of experiment would be one where the monkey notice it send, you know, a state like I need to attend to the left versus the right and ask whether the way it's show patterns that, you know, maybe animal can be put in a state where it knows when and where there are all of your curves.

[01:04:24]
And you know, if you test the idea that maybe the animal has the ability to control the waves and propagate. And she, there are occasions to maximize their detection capability, but I think we're Just sort of archivists reasoning this, assuming the simplest model 1st, I think we'd be really excited.

[01:04:41]
We're also looking at whether there are ordination with the discordant prosecutable areas. So if it waves in a particular state, if you want, it is also occur if you do it in, in an empty, you know, you know, wait, it's a line to Cross a. So this is where the things that we're thinking about in terms of how this might be organized in a larger scale.

[01:05:02]
Thank you. Ok, I have a bunch of questions that we can, you know, turn our back for the sake of time so that people can move on to Jam classes or whatever. Maybe we can collectively thank John for a really great seminar, According chatting like Well, thank you all for having me.

[01:05:26]
It's been a real pleasure to Get to that. This is the 1st time I've done it a little bit rough, totally, somewhat late. I was going, it was clear. All right, I look forward to talking to several of you in the near future and Ok, bye.