WEBVTT

00:00:03.537 --> 00:00:10.497
This is the Convergent Science Network podcast. Leading researchers in the domain

00:00:10.497 --> 00:00:16.737
of neuroscience, brain theory and technology are interviewed by Paul Verschoor and Tony Prescott.

00:00:21.137 --> 00:00:25.757
This is Paul Verschoor with the Convergent Science Network podcast with my colleague

00:00:25.757 --> 00:00:32.977
Tony Prescott. And we're here today with Neil Burgess, who is a speaker in our

00:00:32.977 --> 00:00:35.077
10th edition BCBT Summer School.

00:00:35.437 --> 00:00:41.497
And Neil was giving an overview of his work on spatial cognition,

00:00:41.797 --> 00:00:47.677
also about how to link neural dynamics with advanced psychological function,

00:00:47.877 --> 00:00:49.857
like our knowledge of space.

00:00:49.857 --> 00:00:55.717
And Neil, what you also presented very much in your description of your work

00:00:55.717 --> 00:01:01.817
is this close coupling between theoretical approaches, modeling work, and experimental work.

00:01:02.077 --> 00:01:09.277
So where did this combination of these two methods, if you want, originate?

00:01:10.117 --> 00:01:15.157
Well, I was originally a computational person. Having done theoretical physics,

00:01:15.377 --> 00:01:23.657
I started to want to do computational models of memory and other forms of cognition at the level of neurons.

00:01:26.157 --> 00:01:31.737
And well, my PhD supervisor, theoretical physicist called Mike Moore,

00:01:31.877 --> 00:01:35.337
said, you know, if you're going in this direction, it's an experimental subject

00:01:35.337 --> 00:01:39.657
and you need to get involved in the experiments, and I'm always grateful for that advice.

00:01:40.877 --> 00:01:47.537
There would be no point sticking to mathematically tractable models if they

00:01:47.537 --> 00:01:49.617
had nothing to do with the biological reality,

00:01:49.877 --> 00:01:54.417
and so it seems very important to have your models directly address experiment

00:01:54.417 --> 00:01:59.137
and equally for your experiments to be theoretically well grounded to answer

00:01:59.137 --> 00:02:01.677
particular hypotheses that make sense.

00:02:01.677 --> 00:02:07.117
And so experiment and theory should always work together.

00:02:07.817 --> 00:02:12.437
But now, how many of the models you have generated in that period have you actually

00:02:12.437 --> 00:02:16.497
completely rejected and changed your mind about and taken a completely different direction?

00:02:21.297 --> 00:02:29.537
That's a good question. I think that typically you're interested in a phenomenon

00:02:29.537 --> 00:02:34.177
and you continue to follow that even when experiments or models don't work.

00:02:34.677 --> 00:02:39.477
And so if you're interested in explaining that phenomenon, your models might

00:02:39.477 --> 00:02:48.857
change and indeed your experiments might. so in terms of all that rejection um.

00:02:51.568 --> 00:02:57.228
I'm i'm struggling so i started off uh doing um kind of hopfield model type

00:02:57.228 --> 00:03:02.268
models of memory and i was working with a psychologist graham hitch who was

00:03:02.268 --> 00:03:04.188
interested in memory for serial order,

00:03:05.408 --> 00:03:09.828
and it seemed that although you can remember sequences of things in hopfield

00:03:09.828 --> 00:03:14.928
models by associating one pattern of activity to the next in in human uh working

00:03:14.928 --> 00:03:19.408
memory for serial order Often you make paired transpositions of items.

00:03:19.548 --> 00:03:23.988
So instead of A, B, C, D, you remember, or you output A, C, B, D.

00:03:24.288 --> 00:03:27.708
And it's very hard to get that kind of error in a straightforward Hopfield model.

00:03:27.868 --> 00:03:30.848
And so I started using competitive queuing type models.

00:03:31.068 --> 00:03:34.588
And I haven't really looked back on the Hopfield type models,

00:03:34.748 --> 00:03:41.428
although I still believe that the attractor network idea is key to how much

00:03:41.428 --> 00:03:44.628
of the brain works, including hippocampal area CA3.

00:03:45.328 --> 00:03:49.768
I haven't really gone back to those models as just a pure attractive model of memory.

00:03:50.828 --> 00:03:54.128
Right. So then you started out by saying,

00:03:54.188 --> 00:03:58.688
by pointing to this Wang and Simmons experiment that makes the point that actually

00:03:58.688 --> 00:04:03.848
there are a number of cues or a number of streams of information that come together

00:04:03.848 --> 00:04:05.908
in our understanding of space.

00:04:06.248 --> 00:04:11.648
So why do you think that that experiment is sort of critical as a starting point of the analysis?

00:04:13.828 --> 00:04:17.028
Well, it's just one experiment. I don't think it's critical,

00:04:17.228 --> 00:04:26.908
but it very nicely presents the fact that you can remember where something is in many different ways.

00:04:27.128 --> 00:04:31.788
And so it's sometimes good to start with that example, so that even though I

00:04:31.788 --> 00:04:35.508
then might drone on about hippocampal place cells for a long time,

00:04:35.848 --> 00:04:40.708
the audience already knows that that's only one kind of strategy for remembering

00:04:40.708 --> 00:04:44.888
where something is and that other parts of the brain would be doing equally

00:04:44.888 --> 00:04:48.948
useful things that can help you remember stuff in different ways.

00:04:49.668 --> 00:04:55.548
So of these different possible ways to deal with space, as a concept,

00:04:55.808 --> 00:05:01.048
how would you then define the way the hippocampus solves the problem of space?

00:05:04.508 --> 00:05:11.208
Well, I mean, all of that part of the model is very much driven by the place

00:05:11.208 --> 00:05:16.308
cell phenomenon and surprisingly single neurons encoding for single places.

00:05:16.568 --> 00:05:21.408
I mean it's a remarkable sort of one-to-one mapping aside from the fact that

00:05:21.408 --> 00:05:26.248
different cells will fire in different environments and provides a very easy to use code.

00:05:26.288 --> 00:05:31.928
You could associate things that happen in places with the activity of the place

00:05:31.928 --> 00:05:34.888
cell that fires in that place very easily or you could,

00:05:36.219 --> 00:05:39.039
having found a place cell that fires in a given place it could

00:05:39.039 --> 00:05:42.059
be that its firing rate is modulated by some other factor

00:05:42.059 --> 00:05:44.979
that happens in that place and that's just a very simple starting point

00:05:44.979 --> 00:05:48.219
for thinking about a memory system

00:05:48.219 --> 00:05:51.239
for location yeah i think

00:05:51.239 --> 00:05:54.799
going back to this question about parallel representations a lot

00:05:54.799 --> 00:05:58.859
of the hippocampal models and i think yours start from this assumption that

00:05:58.859 --> 00:06:04.099
the hippocampus represents one solution to the question of where something is

00:06:04.099 --> 00:06:09.179
and but an alternative Bayesian view might be you'd represent a probability

00:06:09.179 --> 00:06:11.459
distribution across multiple solutions.

00:06:11.819 --> 00:06:14.339
I mean, is that something that you entertain as a possibility?

00:06:17.559 --> 00:06:27.099
Yes. So what's been lacking is direct experimental evidence that you might be

00:06:27.099 --> 00:06:31.359
weighing up the different alternatives and seeing place cells firing for both

00:06:31.359 --> 00:06:32.219
different alternatives.

00:06:32.219 --> 00:06:35.819
But in many situations, you can see, for example, as I showed,

00:06:35.879 --> 00:06:40.579
the single firing field of a play cell becoming bimodal when you stretch the

00:06:40.579 --> 00:06:41.419
environment, for example.

00:06:41.719 --> 00:06:43.439
And it's possible that you could

00:06:43.439 --> 00:06:46.579
think of that as part of a system that's trying to estimate location.

00:06:46.919 --> 00:06:51.239
And because it's been predominantly taking evidence from distances to boundaries,

00:06:51.379 --> 00:06:52.879
it's now got a hypothesis that

00:06:52.879 --> 00:06:56.679
has two peaks in it and that you are indeed representing a distribution.

00:06:57.359 --> 00:07:01.379
And in the experiment that I showed where people had to remember where something was,

00:07:02.219 --> 00:07:09.839
The population vector overlap with the stored place cell representation captures

00:07:09.839 --> 00:07:17.639
well the distribution of responses across subjects in that experiment,

00:07:17.879 --> 00:07:20.459
rather than indicating a single location.

00:07:22.019 --> 00:07:26.579
Yeah, but I think Tony's question is also annoying, right?

00:07:26.639 --> 00:07:31.659
Because in some sense, Tony is saying like, well, sort of this spatial cognition

00:07:31.659 --> 00:07:35.179
can happen in many different ways, in many different areas of the brain,

00:07:35.399 --> 00:07:40.979
it becomes a little bit like wishy-washy, like, okay, it's sort of not such

00:07:40.979 --> 00:07:44.139
a unique feature then of a single structure like the hippocampus.

00:07:45.330 --> 00:07:48.890
That's how I take that question. If it's just Bayesian, we integrate multiple

00:07:48.890 --> 00:07:52.530
factors, and so then why worry about hippocampus?

00:07:52.590 --> 00:07:57.010
But I still believe that hippocampus is making a unique contribution to solving that problem.

00:07:57.170 --> 00:08:00.150
Well, I would agree with that. But the question is that, I mean,

00:08:00.190 --> 00:08:03.110
so I think it's definitely true that we have at least two solutions.

00:08:03.430 --> 00:08:07.770
But I'm here to make Neil's life difficult. Just to clarify my position.

00:08:08.410 --> 00:08:12.250
There's a striatal solution to the navigation problem.

00:08:12.790 --> 00:08:17.070
There's something in the hippocampus. There may well be other ways of navigating,

00:08:17.430 --> 00:08:21.730
certainly towards sort of nearby targets and so on.

00:08:23.230 --> 00:08:29.270
So you can almost say, well, in hippocampus, we might be representing multiple

00:08:29.270 --> 00:08:34.210
solutions, and then the decision gets made in some other part of the brain about which way we would go.

00:08:34.410 --> 00:08:39.930
And there are models of hippocampus that are consistent with the idea of representing multiple hypotheses.

00:08:41.270 --> 00:08:44.310
I want Neil to tell me now which one it is. Yes.

00:08:46.730 --> 00:08:51.210
Well i mean as as i showed um there are

00:08:51.210 --> 00:08:54.210
some things that we do know about the brain and this packard

00:08:54.210 --> 00:08:57.190
and magor experiment was again it was

00:08:57.190 --> 00:08:59.850
only a summary of lots of things that people knew up to that point but it

00:08:59.850 --> 00:09:04.070
was very nice clear expression of the fact that different solutions

00:09:04.070 --> 00:09:08.850
could um seem to be supported by different parts of the brain and you could

00:09:08.850 --> 00:09:13.290
actually inject anesthetic into the two different parts of the brain and see

00:09:13.290 --> 00:09:18.630
the different strategies being expressed in behavior so um in answer to tony's

00:09:18.630 --> 00:09:22.350
question i think well and indeed yours that um.

00:09:23.510 --> 00:09:26.170
If i want to characterize the hippocampal contribution to these

00:09:26.170 --> 00:09:29.110
kind of spatial memory tasks you know it it seems to

00:09:29.110 --> 00:09:32.090
be representing a location relative to

00:09:32.090 --> 00:09:35.230
the environment probably as a conjunction of many kinds of cues some

00:09:35.230 --> 00:09:39.210
of which are distances to boundaries in different directions whereas

00:09:39.210 --> 00:09:42.110
other parts of the brain might be remembering more uh you

00:09:42.110 --> 00:09:45.150
know i turned left at the shoe shop or a sequence of turns on

00:09:45.150 --> 00:09:48.590
a well-known route or in other parts of the brain still

00:09:48.590 --> 00:09:51.850
you know a visual visual snapshot matching uh

00:09:51.850 --> 00:09:55.150
this is this looks like where i was before and all

00:09:55.150 --> 00:09:59.850
of these things are happening in parallel and it is an interesting question

00:09:59.850 --> 00:10:04.570
what how they get combined who chooses what and indeed on the input there's

00:10:04.570 --> 00:10:07.890
different kinds of evidence coming in and they are probably waiting in a bayesian

00:10:07.890 --> 00:10:12.310
way with more things that seem to to be more stable over time or more certain,

00:10:12.370 --> 00:10:15.450
being given a stronger weight than things that seem more variable.

00:10:15.970 --> 00:10:20.590
But would it be fair then to say that the hippocampus really is critical in

00:10:20.590 --> 00:10:23.370
generating this allocentric representation of space?

00:10:25.354 --> 00:10:29.534
Yes, I would think so, yeah. But in many of these experiments,

00:10:29.834 --> 00:10:31.394
they're a memory experiment.

00:10:31.514 --> 00:10:33.494
You have to remember a location and go there.

00:10:33.634 --> 00:10:36.674
And we know that the hippocampus is critical for many forms of memory.

00:10:36.874 --> 00:10:40.634
And so in a spatial memory experiment, yes, if you inactivate it,

00:10:40.734 --> 00:10:43.994
then often the person or animal will not perform very well.

00:10:44.194 --> 00:10:47.194
So there's good evidence that the hippocampus is responsible. responsible and

00:10:47.194 --> 00:10:50.054
in that pakada magore experiment you could see that

00:10:50.054 --> 00:10:52.894
it was particularly responsible for one of the two available

00:10:52.894 --> 00:10:56.654
strategies for that task yeah the general hypothesis

00:10:56.654 --> 00:11:00.994
that you're following then is that the there's a population response in hippocampus

00:11:00.994 --> 00:11:06.294
of of play cells which in some way represent the brain's best guess as to where

00:11:06.294 --> 00:11:10.794
it is in the world in an alec-centric coordinate frame in those tasks yes yeah

00:11:10.794 --> 00:11:15.414
okay we're done so now we we have an hypothesis

00:11:15.714 --> 00:11:18.194
about what the hippocampus is doing.

00:11:18.514 --> 00:11:21.114
And it gives you this, an ellicentric representation of space.

00:11:21.174 --> 00:11:24.574
So in world coordinates, which is, which is a fantastic accomplishment.

00:11:24.774 --> 00:11:27.334
And now we have to try to understand how this is done.

00:11:27.674 --> 00:11:32.154
Right. And, and then you build very much your, your theory or model on that

00:11:32.154 --> 00:11:37.034
around, at least in the, in your presentation now around this notion of the boundary vector cells.

00:11:37.174 --> 00:11:39.894
This is really the starting point of your whole analysis, right?

00:11:39.954 --> 00:11:45.054
So, so why, why do you feel that this notion of boundary vector cells is then

00:11:45.054 --> 00:11:49.234
so decisive to understand this ability to generate allocentric representation of space?

00:11:50.574 --> 00:11:54.534
Well, first of all, the head direction cells, I think, are fundamental because

00:11:54.534 --> 00:11:58.114
this overall orientation, sense of orientation, is important for everything

00:11:58.114 --> 00:12:00.274
that comes subsequent to that.

00:12:01.054 --> 00:12:09.154
But taking that as granted, then you need some distances from some things to know where you are.

00:12:10.221 --> 00:12:16.461
And really, I think that if you were to imagine an animal with a sort of range

00:12:16.461 --> 00:12:21.181
finder, looking at the distances to stuff around it in full 360 degrees,

00:12:21.521 --> 00:12:24.821
you know, obviously, the biggest contribution will be to either things that

00:12:24.821 --> 00:12:27.021
are very nearby or things that are quite extended.

00:12:27.501 --> 00:12:32.301
And so topographical features of the environment that are extended are going

00:12:32.301 --> 00:12:33.801
to dominate that kind of representation.

00:12:34.581 --> 00:12:38.361
And I think that that's really the boundary vector cells.

00:12:38.361 --> 00:12:41.101
Cells we heard also and and

00:12:41.101 --> 00:12:44.821
i talked about object vector cells and we heard from them from edvard

00:12:44.821 --> 00:12:47.661
moser you know the

00:12:47.661 --> 00:12:50.901
boundary vector cells would do actually respond to uh

00:12:50.901 --> 00:12:54.301
small objects just they have a very small firing

00:12:54.301 --> 00:12:57.321
field is provoked by a very small object and so

00:12:57.321 --> 00:13:00.641
they're sort of responding like a a range

00:13:00.641 --> 00:13:04.021
finder tuned in a particular direction allocentric direction

00:13:04.021 --> 00:13:07.301
presumably governed by the head direction cells and and

00:13:07.301 --> 00:13:10.101
you know with once you're oriented the next

00:13:10.101 --> 00:13:14.061
thing you need is distance and and these are range finders

00:13:14.061 --> 00:13:16.841
tuned to directions and they seem to

00:13:16.841 --> 00:13:20.061
provide the simplest explanation

00:13:20.061 --> 00:13:25.701
for the sort of broad qualities of the firing fields of place cells when you

00:13:25.701 --> 00:13:29.141
put an animal in different environments environments that differ in shape and

00:13:29.141 --> 00:13:33.981
size but not differ in in texture and odor and and so on enough to make different

00:13:33.981 --> 00:13:37.761
play cells fire in which case you wouldn't be able to try and analyze these different kinds of.

00:13:38.948 --> 00:13:42.708
So then you're saying, look, I have a heading vector. This gives me,

00:13:42.788 --> 00:13:46.688
if you want, that's my compass. That's my directional system.

00:13:47.028 --> 00:13:51.368
And now I glue to that some sensory feature.

00:13:52.188 --> 00:13:55.768
Might be somatosensory. It might be auditory. It might be visual.

00:13:56.248 --> 00:14:02.288
And that now gives me, let's say, a reference in space of how that compass coordinate

00:14:02.288 --> 00:14:05.228
maps onto that specific environment.

00:14:05.228 --> 00:14:08.908
Environment well i should say that in most normal environments

00:14:08.908 --> 00:14:11.828
there'll be a you know a broad range of different

00:14:11.828 --> 00:14:14.688
local cues also which would help you in a very direct way

00:14:14.688 --> 00:14:17.848
to localize yourself and they are probably also important to

00:14:17.848 --> 00:14:20.848
play cells but in most experiments most people

00:14:20.848 --> 00:14:24.048
do they try to control for those cues so that they're only

00:14:24.048 --> 00:14:27.688
the cues that that they're aware of remaining like

00:14:27.688 --> 00:14:30.668
the boundary of the environment these distant orientation cues for the

00:14:30.668 --> 00:14:34.388
head iteration cells and so um given

00:14:34.388 --> 00:14:37.808
that you've tried to remove reliable local queues

00:14:37.808 --> 00:14:41.148
also usually the floor is rotated between trials then

00:14:41.148 --> 00:14:44.248
what you have left you have distances to to

00:14:44.248 --> 00:14:50.008
extended things right but now the head direction system depends on path integration

00:14:50.008 --> 00:14:56.748
path integration is noisy because it's an integrator um so isn't that a little

00:14:56.748 --> 00:15:00.688
bit of an obstacle for for the such which is to keep on working reliably over

00:15:00.688 --> 00:15:01.928
extended periods of time. Yes, indeed.

00:15:02.068 --> 00:15:07.728
So that's an interesting thing that I didn't talk about, but exactly the same

00:15:07.728 --> 00:15:11.908
concerns about how you translate between egocentric and allocentric in terms

00:15:11.908 --> 00:15:16.168
of place cells and sensory inputs, which have to be egocentric.

00:15:16.948 --> 00:15:18.068
Applies to head direction cells.

00:15:18.228 --> 00:15:21.968
So a head direction cell will fire whenever the animal's facing,

00:15:22.108 --> 00:15:24.928
let's say, north. I don't actually mean compass north.

00:15:25.848 --> 00:15:30.808
Wherever it is in the environment. And so actually in all experimental situations

00:15:30.808 --> 00:15:31.908
when they've been recorded.

00:15:33.664 --> 00:15:37.684
That tuning is parallel across the environment, but if they were simply responding

00:15:37.684 --> 00:15:40.484
to sensory cues, you would see parallax, which you do not see.

00:15:40.704 --> 00:15:43.784
It is an allocentric response, the head direction cells. And so again,

00:15:43.884 --> 00:15:47.264
you need to have this translation mechanism between egocentric sensory input

00:15:47.264 --> 00:15:48.844
and this allocentric response.

00:15:48.844 --> 00:15:53.264
And so we think probably retrosplenial cortex, as in the model that I explained,

00:15:53.524 --> 00:15:56.804
has to do that same job for the head direction cells so that they can be reliably

00:15:56.804 --> 00:16:01.584
anchored to sensory input to prevent the accumulation of error that you would

00:16:01.584 --> 00:16:06.804
inevitably get if you're just integrating angular acceleration. Right.

00:16:07.224 --> 00:16:12.584
But that is the sensory cue that you link the heading vector to.

00:16:13.704 --> 00:16:18.424
That in itself might already be an invariant representation of some sensory state.

00:16:18.584 --> 00:16:21.404
It might not be necessarily egocentric. right if

00:16:21.404 --> 00:16:24.244
we go through some some perceptual hierarchy and if

00:16:24.244 --> 00:16:27.044
i climb up that hierarchy i might get to more invariant representations

00:16:27.044 --> 00:16:30.344
of space maybe i have a very abstract representation say

00:16:30.344 --> 00:16:34.044
just okay this is a wall i'm not

00:16:34.044 --> 00:16:36.964
interested anymore in its orientation or its color or

00:16:36.964 --> 00:16:39.764
its size a wall so so how would that notion

00:16:39.764 --> 00:16:43.664
of invariant representations of the sensory feature then actually

00:16:43.664 --> 00:16:47.324
be a challenge to that to that model well i

00:16:47.324 --> 00:16:50.044
think it's um the invariance we're

00:16:50.044 --> 00:16:52.904
talking about uh is exactly you know the head direction

00:16:52.904 --> 00:16:55.664
cells show that very nicely they're at

00:16:55.664 --> 00:17:01.624
the top of the hierarchy for direction they're invariant to all of the parallax

00:17:01.624 --> 00:17:05.524
and sensory stuff that happens as you move around and they extract what would

00:17:05.524 --> 00:17:11.324
be a compass you know and and uh the rest of the system is probably built on

00:17:11.324 --> 00:17:15.324
that and now if it would replace the heading vector with a movement vector?

00:17:15.564 --> 00:17:18.404
Would that change your approach very much?

00:17:24.044 --> 00:17:29.004
I'm not sure what you mean. As opposed to integrating, let's say, my rotations in space,

00:17:29.124 --> 00:17:34.624
I could also say I just interpolate between subsequent movements,

00:17:34.724 --> 00:17:40.244
and this gives me a vector that I might then use to get to some sort of LSM

00:17:40.244 --> 00:17:41.164
representation of space.

00:17:42.144 --> 00:17:45.484
Well, so there's two separate things here. One is how do the head direction

00:17:45.484 --> 00:17:49.624
cells fire to encode the orientation of the head. If you then move on to path

00:17:49.624 --> 00:17:53.344
integration and perhaps what drives grid cells to fire, then you have to integrate

00:17:53.344 --> 00:17:54.964
movement rather than head direction.

00:17:55.184 --> 00:17:59.964
So you need to know which direction you've moved in, not which direction your head is facing.

00:18:00.244 --> 00:18:03.384
Exactly. So would that make a big difference to your proposal?

00:18:05.344 --> 00:18:10.344
Not for the boundary vector cells, I don't think. Because usually the head of

00:18:10.344 --> 00:18:14.764
the rat is... Because the boundary vector cell model of place cell firing is

00:18:14.764 --> 00:18:17.244
not path integration. It's just feed-forward sensory input.

00:18:17.724 --> 00:18:21.464
What direction from the head is the boundary feed forward.

00:18:21.564 --> 00:18:24.844
So there's not an issue there. When you come on to how you do path integration,

00:18:25.164 --> 00:18:28.624
how you integrate your own bodily movement to estimate where you are,

00:18:28.724 --> 00:18:30.544
which may well happen through grid cells,

00:18:31.044 --> 00:18:34.684
then indeed you need to know your movement direction, your movement vector,

00:18:34.804 --> 00:18:37.084
not your heading vector. That's right. Perfect, yeah.

00:18:37.444 --> 00:18:41.584
And so there's various hypotheses as to where this signal comes from because

00:18:41.584 --> 00:18:44.524
it can't be just the head direction. That's right.

00:18:45.004 --> 00:18:49.004
Multiplied by running speed. has to be movement direction. Would it be your grid cells then?

00:18:51.004 --> 00:18:55.284
No, what we're talking about are what are the inputs that would go to something

00:18:55.284 --> 00:18:59.084
like grid cells that is the movement signal that is integrated.

00:19:00.704 --> 00:19:06.904
If I would just look at subsequent grid cell responses, I could infer a movement vector.

00:19:07.104 --> 00:19:09.884
Yes, you could do. So that's what I was after. Yes,

00:19:09.884 --> 00:19:12.664
so I think maybe this question came up

00:19:12.664 --> 00:19:16.124
with Edvard Moser's talk that whether you could

00:19:16.124 --> 00:19:19.424
look at it either way if you think grid cells know uh

00:19:19.424 --> 00:19:22.124
how to fire because they're doing path integration you need

00:19:22.124 --> 00:19:25.184
an input which is a movement vector or you could say well

00:19:25.184 --> 00:19:28.564
grid cells fire as they do in which case a movement

00:19:28.564 --> 00:19:32.384
vector could be an output yes that's it yes exactly indeed it could be but then

00:19:32.384 --> 00:19:37.504
you're still left with a question of how the grid cells fired in that way in

00:19:37.504 --> 00:19:43.224
the first place right the um so the the activation of the play cells you have

00:19:43.224 --> 00:19:46.564
of being driven by the boundary vector cells.

00:19:46.884 --> 00:19:52.124
And so it's basically a thresholded sum of the inputs from the boundary cells.

00:19:54.684 --> 00:20:00.724
Clearly, you could choose the weighted sum of the inputs as just the first thing to try.

00:20:00.904 --> 00:20:04.324
Are there other motivations for doing it that way, for example,

00:20:04.324 --> 00:20:08.984
from environment warping experiments that say that it's the weighted average,

00:20:09.164 --> 00:20:10.224
it's the best thing to use?

00:20:11.924 --> 00:20:14.764
Well, it was the simplest thing, the first thing to try.

00:20:14.764 --> 00:20:17.724
But in some experiments as you um expand the

00:20:17.724 --> 00:20:20.804
box you have a place field which slowly reduces in

00:20:20.804 --> 00:20:24.044
firing rate and disappears right which implies a

00:20:24.044 --> 00:20:26.924
sum with a threshold you know as the

00:20:26.924 --> 00:20:29.824
over initially overlapping inputs from opposing

00:20:29.824 --> 00:20:35.144
walls uh move apart then the the bit you've got left slowly falls below threshold

00:20:35.144 --> 00:20:40.324
and the cell stops firing so there's there's good reason to use a thresholded

00:20:40.324 --> 00:20:48.244
sum there's a reason for a threshold and um we really did see um in some cases individual,

00:20:48.784 --> 00:20:51.464
sub peaks being moved apart which means you have

00:20:51.464 --> 00:20:54.184
to sum the two rather multiple if you multiply them then

00:20:54.184 --> 00:20:57.484
you get for example a bayesian you know average position

00:20:57.484 --> 00:21:00.104
in between the two peaks as they well this is what i was

00:21:00.104 --> 00:21:02.984
getting at is there support for multiple hypotheses in some

00:21:02.984 --> 00:21:05.724
of those well um just to

00:21:05.724 --> 00:21:08.784
finish answering question there's there's clearly adding these

00:21:08.784 --> 00:21:12.064
inputs from the different walls and having a threshold was the

00:21:12.064 --> 00:21:18.264
most obvious thing to do given all the things we've said um so if you were trying

00:21:18.264 --> 00:21:24.244
to combine um hypotheses probabilistically you would probably multiply and that's

00:21:24.244 --> 00:21:29.544
not what we saw right um it's interesting because of these experiments by um,

00:21:30.264 --> 00:21:36.104
galasell and cheng and also by uh cartwright and collett um john o'keefe in

00:21:36.104 --> 00:21:39.644
this original stretchy box experiment was expecting to see.

00:21:40.484 --> 00:21:42.284
Cells, place cells, firing.

00:21:43.944 --> 00:21:50.664
In the constant ratio of the distances across the box, rather than the sub-peaks

00:21:50.664 --> 00:21:55.284
being pulled apart and sticking to the fixed absolute distance from the walls

00:21:55.284 --> 00:21:57.724
as they were pulled apart. And so...

00:21:58.844 --> 00:22:01.644
If what he'd been expecting had happened, that would have been an argument for

00:22:01.644 --> 00:22:06.024
multiplying and having the combined estimate of two probabilities distributions,

00:22:06.644 --> 00:22:07.684
but that isn't what we saw.

00:22:07.944 --> 00:22:13.904
But it sounds like what the experiment says is a bit more sophisticated than the averaging.

00:22:15.564 --> 00:22:18.444
So there might be something to look at there in

00:22:18.444 --> 00:22:21.524
the future in terms of how in this inconsistent

00:22:21.524 --> 00:22:25.204
consistent environment that's you know where uh the

00:22:25.204 --> 00:22:28.044
boundary effect cells don't all point to

00:22:28.044 --> 00:22:30.984
the same place in space then you might start to represent

00:22:30.984 --> 00:22:33.844
multiple hypotheses in terms of place cell activity

00:22:33.844 --> 00:22:36.904
well i don't i i don't know if there's

00:22:36.904 --> 00:22:40.664
evidence for that i think there is definitely evidence for um

00:22:40.664 --> 00:22:43.584
hypotheses about what happens in

00:22:43.584 --> 00:22:46.464
a place you know place cell a place cell

00:22:46.464 --> 00:22:49.864
might fire in the same place trial after trial but its

00:22:49.864 --> 00:22:52.744
firing rate might vary quite strongly according to

00:22:52.744 --> 00:22:56.424
other hypotheses to do with sensory stimuli

00:22:56.424 --> 00:22:59.764
that are present whether objects are nearby smells how

00:22:59.764 --> 00:23:02.464
fast you're running so I think there's a lot of

00:23:02.464 --> 00:23:10.184
orthogonal hypotheses to place which also modulate the firing rate but it more

00:23:10.184 --> 00:23:16.824
like a gain field representation okay so the actual firing field itself uh isn't

00:23:16.824 --> 00:23:20.324
obviously doing something nice and bayesian about estimating location.

00:23:21.284 --> 00:23:25.284
Uh when you do these weird experiments when you deform the environment which

00:23:25.284 --> 00:23:30.424
are rather unnatural but they probably are combining the cues to them such as

00:23:30.424 --> 00:23:34.864
we haven't talked about the path integration input and the sensory input are

00:23:34.864 --> 00:23:38.824
probably being combined in a more bayesian way yeah and also as you say maybe

00:23:38.824 --> 00:23:40.924
there's something happening in time you know that,

00:23:41.484 --> 00:23:45.764
you can swap between hypotheses over time and explore different interpretations.

00:23:45.764 --> 00:23:50.464
Well, actually, on that point, if you do this stretching experiment and you

00:23:50.464 --> 00:23:56.144
get a single place field stretching into two place fields as you've expanded the box,

00:23:56.284 --> 00:24:02.824
then you notice that each place field fires a bit more according to running direction. So the...

00:24:04.128 --> 00:24:08.988
The place field that appears to be attached to the wall behind the animal fires

00:24:08.988 --> 00:24:12.448
a bit more, so that when you're running in one direction, one peak will be higher,

00:24:12.508 --> 00:24:14.428
and when you're running in the other direction, the other peak will be higher,

00:24:14.508 --> 00:24:17.848
which we interpreted as being this path integration input,

00:24:18.068 --> 00:24:25.048
adding to this environmental sensory input, and being stronger if you just run from a wall.

00:24:25.248 --> 00:24:28.608
Obviously, you have a good estimate of how far away it is from path integration,

00:24:28.848 --> 00:24:32.088
so that peak gets a bit higher because it's got a stronger path integration

00:24:32.088 --> 00:24:35.728
input but down the peak that's attached to the opposite wall that you're running towards.

00:24:36.308 --> 00:24:39.288
Yeah, yeah. But I find it very interesting.

00:24:39.348 --> 00:24:45.188
You guys seem so, let's say, motivated to interpret this in Bayesian terms.

00:24:45.408 --> 00:24:49.828
Well, if you look at the dynamics of CA3, CA1, where the real action in the

00:24:49.828 --> 00:24:54.308
end happens, it's also highly competitive, and in the end, it's relatively sparse.

00:24:54.388 --> 00:24:57.128
You can also think about it much more as, let's say.

00:24:57.708 --> 00:25:02.988
The endpoint selector that sits on top of some Bayesian integrator as opposed

00:25:02.988 --> 00:25:06.348
to being a Bayesian integrator itself because you just don't have the dynamics

00:25:06.348 --> 00:25:10.188
for it because it's rather sort of selective,

00:25:10.468 --> 00:25:14.528
rather sparse, and these are not the features you want in your Bayesian integrator.

00:25:14.708 --> 00:25:18.928
So is it not fair to place that bit outside of the hippocampus, gentlemen?

00:25:19.048 --> 00:25:25.828
No, I think that's fine. As I said initially to Tony, there isn't good evidence

00:25:25.828 --> 00:25:28.968
of multiple hypotheses being entertained at the same time.

00:25:29.148 --> 00:25:35.128
It would be a good way for the system to work but it may well not do that.

00:25:35.288 --> 00:25:40.748
And then can't we not explain the elongation of the place field also in terms

00:25:40.748 --> 00:25:47.828
of a perceptual learning process because the animal starts also carve out specific

00:25:47.828 --> 00:25:49.768
trajectories through that space.

00:25:49.988 --> 00:25:52.828
It's just not moving randomly, right? And the cells you

00:25:52.828 --> 00:25:55.628
measure from is probably also on trajectories the animal

00:25:55.628 --> 00:25:58.788
will visit more frequently than other trajectories so it

00:25:58.788 --> 00:26:01.988
will actually be moving more often from that specific

00:26:01.988 --> 00:26:08.328
initial position to a final position and therefore expose itself more to the

00:26:08.328 --> 00:26:12.648
specific sensory features that are already associated with that place field

00:26:12.648 --> 00:26:18.268
and therefore strengthen the specific intra-place field associations and then,

00:26:18.328 --> 00:26:19.728
if you want, elongating its response.

00:26:20.588 --> 00:26:24.708
Well, there are interesting trajectory effects on place cell firing,

00:26:24.828 --> 00:26:28.008
but they're rather different than these effects in this unusual situation where

00:26:28.008 --> 00:26:30.388
we change the shape and size of the environment.

00:26:30.588 --> 00:26:34.748
I mean, I think that's an unusual experimental manipulation that shows us something about the inputs.

00:26:35.248 --> 00:26:40.608
However, the trajectory effects are interesting so that after... So if the...

00:26:41.911 --> 00:26:44.831
The animal's always running in the same direction. You see that firing fields

00:26:44.831 --> 00:26:49.331
tend to elongate backwards a little bit along the track. So my anchor, Mater, noticed this.

00:26:51.771 --> 00:26:56.911
But that's been interpreted as perhaps the play cell beginning to fire in expectation

00:26:56.911 --> 00:27:00.211
of getting to its, quote, true place, original place.

00:27:00.571 --> 00:27:06.851
And there's some interesting theoretical work by Matt Botvinick and some co-authors

00:27:06.851 --> 00:27:14.191
suggesting that these play cells might encode a successor representation for

00:27:14.191 --> 00:27:14.951
reinforcement learning,

00:27:15.151 --> 00:27:20.071
which is an idea from Peter Diane in the 90s, that if you were to tweak your

00:27:20.071 --> 00:27:25.991
state representation to actually include the likelihood of eventually ending

00:27:25.991 --> 00:27:28.391
up at that state, even when you're at other states,

00:27:28.591 --> 00:27:37.191
then that enables you to estimate future reward at a given location much more efficiently.

00:27:37.511 --> 00:27:40.271
And so it's possible that if you have repeated trajectory, trajectories,

00:27:40.511 --> 00:27:46.651
then you can build a little bit of the probability that you now know that if you're here,

00:27:46.711 --> 00:27:51.271
you're likely to end up somewhere else into the state representation so that

00:27:51.271 --> 00:27:57.411
you can then rather more easily estimate likely future reward from a given location

00:27:57.411 --> 00:27:59.191
because you've built in the transition probabilities,

00:27:59.491 --> 00:28:03.551
which are now not uniform because I always run in this direction when I'm in this place.

00:28:03.711 --> 00:28:06.411
So it's the kind of model that reduces the whole brain into the hippocampus.

00:28:07.011 --> 00:28:09.751
Well, no. I mean, it may be that that kind of representation,

00:28:10.191 --> 00:28:13.471
you know, perceptual learning and so on, affects, you know, it could explain

00:28:13.471 --> 00:28:17.031
what happens in lots of parts of the brain, not just hippocampus. Right, okay.

00:28:17.251 --> 00:28:21.271
But now, so in your physiology, so then with that model, you went to look at the physiology,

00:28:21.931 --> 00:28:29.411
and then you show that these sort of vector border cells are to be found both

00:28:29.411 --> 00:28:33.711
in the subiculum and the entorhinal cortex, which is interesting, right?

00:28:33.711 --> 00:28:36.791
Because if you look at the overall loop, we start in entorhinal,

00:28:37.011 --> 00:28:43.591
we go to the dentate, gyrus, CA3, CA2, CA1, then subiculum, and then we go out to entorhinal cortex.

00:28:43.951 --> 00:28:47.611
So isn't it really strange that we find these cells actually at the input stage

00:28:47.611 --> 00:28:49.671
and the output states of that whole loop?

00:28:49.791 --> 00:28:52.831
It's like we're wasting all this circuitry in between to do nothing,

00:28:52.931 --> 00:28:55.451
and we just have our border vector cells.

00:28:55.771 --> 00:29:00.171
It is. It is strange. I mean, it is a loop, and we may be interpreting how it

00:29:00.171 --> 00:29:05.171
works incorrectly when we think of entorhinal as the input and subiculum as the output.

00:29:05.371 --> 00:29:07.791
I mean, there's plenty of connections from subiculum back to entorhinal,

00:29:07.851 --> 00:29:10.791
and presubiculum has very strong connections to entorhinal.

00:29:13.032 --> 00:29:18.552
My colleague Colin Lever is always keen to point out that something that Pat

00:29:18.552 --> 00:29:22.232
Sharp originally noticed, that when you get a complete remapping of place cells

00:29:22.232 --> 00:29:23.972
in CA1 between two environments,

00:29:24.572 --> 00:29:29.992
there's actually spatial responses in subiculum that are nice and stable that

00:29:29.992 --> 00:29:32.012
you can record, and they show no remapping at all.

00:29:32.312 --> 00:29:36.252
And yet the standard model would be that the major input to subiculum is the

00:29:36.252 --> 00:29:42.312
output of CA1, and yet you've got this complete change in CA1 representation, no change in subiculum.

00:29:42.312 --> 00:29:45.352
Them so yes it is a puzzle you know maybe

00:29:45.352 --> 00:29:48.152
subiculum is the input and it goes around to enter

00:29:48.152 --> 00:29:50.952
and then into you know who knows but it is a puzzle

00:29:50.952 --> 00:29:53.652
it's puzzling from the anatomy you wouldn't expect it no but

00:29:53.652 --> 00:29:56.512
it's quite a challenge there's something there that but it

00:29:56.512 --> 00:29:59.292
is a challenge to your model right because you want to say these border vector

00:29:59.292 --> 00:30:03.292
cells are if you want a representational primitive on which i built everything

00:30:03.292 --> 00:30:06.652
else but now we see whatever wherever you want to put the input and the output

00:30:06.652 --> 00:30:11.192
they still sit there well and and what is also true if you you know you You

00:30:11.192 --> 00:30:18.592
could also ask me the sort of functional difficult question would be the place cells are a basis set.

00:30:18.732 --> 00:30:23.132
You know, it could be that boundary vector cell firing responses are made out

00:30:23.132 --> 00:30:27.752
of weighted sums of place cell firing. That could be possible.

00:30:28.752 --> 00:30:33.332
But now, if you look at subiculum and entorhinal, how big a proportion of cells

00:30:33.332 --> 00:30:36.192
in those areas would actually show these properties?

00:30:37.419 --> 00:30:43.119
So I think it's around 20% or so is Colin Lever's best estimate in subiculum,

00:30:43.159 --> 00:30:46.439
and it's a bit less, maybe more like 10%, but still quite common,

00:30:46.479 --> 00:30:50.319
these border cells that Edvard Moser described, i.e.

00:30:50.339 --> 00:30:53.399
The boundary vector cells that fire quite close to the boundary. boundary

00:30:53.399 --> 00:30:56.559
they're quite common there have been a few reported

00:30:56.559 --> 00:31:00.259
that fire at a distance but it's a handful and so that's

00:31:00.259 --> 00:31:06.419
another another strange yeah we that remains to be um the exact relationship

00:31:06.419 --> 00:31:09.779
between the border cells and the boundary vector cells and the object vector

00:31:09.779 --> 00:31:17.219
cells is still not quite clear okay so but then could we argue that in entorhinal cortex there's if you

00:31:17.239 --> 00:31:22.699
want a rank order of cells and in terms of how foundational they are.

00:31:22.819 --> 00:31:25.679
You could argue, well, grid cells might be very foundational.

00:31:27.259 --> 00:31:30.679
Single modality cells in lateral entorhinal cortex might be foundational.

00:31:31.199 --> 00:31:35.319
And now I start to merge them. I start to merge them in your border vector cells.

00:31:35.599 --> 00:31:40.239
So because now I have within entorhinal cortex, I have access to both movement

00:31:40.239 --> 00:31:43.819
vectors, grid cells, and have access to sensory features of the world,

00:31:43.859 --> 00:31:47.739
lateral entorhinal cortex. and I just have a subpopulation of cells that is

00:31:47.739 --> 00:31:50.219
associating the two, and now I have my border vector cells.

00:31:52.719 --> 00:31:59.879
Yes, the grid cells are also a basis set, and you could form any responses out of them.

00:32:00.899 --> 00:32:05.659
So it's a good question, although it's not quite clear what foundational means in your question.

00:32:05.919 --> 00:32:10.719
And so we have looked developmentally. So Francesca Cucci in John O'Keefe's

00:32:10.719 --> 00:32:16.439
lab was looking at this and also Edvard Moser's lab at the same time.

00:32:17.912 --> 00:32:22.092
And what you see during development as pups learn to start to crawl around and

00:32:22.092 --> 00:32:25.312
open their eyes and so on is that the head direction cells are foundational

00:32:25.312 --> 00:32:28.052
in the sense that they seem to be there as soon as you can record them.

00:32:28.832 --> 00:32:32.852
There's also theta rhythmicity as early as you can record them in this situation.

00:32:33.632 --> 00:32:37.132
Then you also see place cell responses as they start to crawl around,

00:32:37.292 --> 00:32:43.212
the rat pups, and they get better with experience over several weeks.

00:32:43.212 --> 00:32:47.912
So there's a slow improvement in the spatial tuning of the play cells.

00:32:48.072 --> 00:32:55.392
Grid cells are not present until four or five developmental days after you see these other cells.

00:32:55.572 --> 00:32:58.592
And so they're not foundational in that sense, it seems.

00:32:59.432 --> 00:33:02.632
And this has led to people suggesting that maybe

00:33:02.632 --> 00:33:05.572
the grid cells require some stable spatial input maybe from

00:33:05.572 --> 00:33:08.332
play cells to wire themselves up to fire

00:33:08.332 --> 00:33:11.092
properly i mean uh the

00:33:11.092 --> 00:33:14.272
head direction cell seems to be a very strong attractor network

00:33:14.272 --> 00:33:18.292
so that the the head direction cell responses are

00:33:18.292 --> 00:33:21.112
mutually coherent with each other even if they are drifting a little

00:33:21.112 --> 00:33:24.952
bit early on in development it may be the same with the grid cells that they're

00:33:24.952 --> 00:33:28.572
mutually connected well so that it's an attractor it's just drifting so much

00:33:28.572 --> 00:33:33.312
relative to the world that you can't ever record it that's a possibility and

00:33:33.312 --> 00:33:36.692
then you only see it when it gets attached to stable sensory inputs because

00:33:36.692 --> 00:33:38.752
then you can average over movements across the environment.

00:33:39.652 --> 00:33:42.532
But yeah they're all interesting questions and you can try the good

00:33:42.532 --> 00:33:46.512
thing about this field is that you can try to answer them with experiments absolutely

00:33:46.512 --> 00:33:50.692
and the developmental trajectory you're talking about there is based on the

00:33:50.692 --> 00:33:54.212
chemical properties of the cells not not their functional properties no no no

00:33:54.212 --> 00:34:00.392
sorry the function this is recording um uh the the the spiking activity as

00:34:00.492 --> 00:34:05.572
pups first start to move around as they start to explore around the nest.

00:34:05.972 --> 00:34:08.892
Francesco will give a talk about this tomorrow, I think. Right.

00:34:09.512 --> 00:34:14.292
But now the object vector cells, this is the next complexification that also

00:34:14.292 --> 00:34:15.672
Edward talked about yesterday.

00:34:16.932 --> 00:34:21.612
You see them as a further variation on that same theme, or should we now be

00:34:21.612 --> 00:34:24.212
really shocked and say, oh, this is again completely different?

00:34:25.392 --> 00:34:28.792
And oh well they're obviously closely related to

00:34:28.792 --> 00:34:32.512
boundary vector cells in the sense of firing at a allocentric displacement

00:34:32.512 --> 00:34:35.392
vector from something and the only thing that's different is that they

00:34:35.392 --> 00:34:39.172
seem to be specifically tuned to small objects right

00:34:39.172 --> 00:34:41.932
so a boundary vector cell as i said will fire to an extended object but will

00:34:41.932 --> 00:34:44.772
also fire to a smaller object but just with

00:34:44.772 --> 00:34:47.992
a very small firing field because uh obviously it

00:34:47.992 --> 00:34:50.852
only has to move a little way and then it's not in in the receptive

00:34:50.852 --> 00:34:54.552
field for that cell whereas an extended boundary will produce a long stripify

00:34:54.552 --> 00:35:01.252
the object vector cells seem to be different in that they uh don't seem to respond

00:35:01.252 --> 00:35:05.692
to the extended boundaries of the environment they just respond to objects put

00:35:05.692 --> 00:35:10.312
within it now it could be that over time uh.

00:35:12.007 --> 00:35:15.287
There's something to do with the object being novel and the boundaries being

00:35:15.287 --> 00:35:19.507
familiar means that it discriminates because the same object vector cell will

00:35:19.507 --> 00:35:21.527
respond similarly to different objects.

00:35:21.727 --> 00:35:27.027
It's not specific, it seems, to specific objects, but will fire to any small

00:35:27.027 --> 00:35:29.467
object, but not to an extended boundary.

00:35:29.667 --> 00:35:33.707
But so far, the experiments, I don't think, have really been done to see whether

00:35:33.707 --> 00:35:37.407
if you put a novel extended object in that looks a bit like a boundary,

00:35:37.527 --> 00:35:39.887
whether those cells will fire or not. Exactly, that's the question, right?

00:35:39.907 --> 00:35:43.067
When does an object become a boundary? The reverse experiment has sort of been

00:35:43.067 --> 00:35:50.347
done in that Bruno Poussey's group had showed that some wine bottles put inside

00:35:50.347 --> 00:35:53.127
one of these boxes, when you're calling place cells,

00:35:53.327 --> 00:35:56.147
typically don't cause much place cell firing.

00:35:56.327 --> 00:36:00.847
Although we know now from Jim Nerium's work that a small number of place cells

00:36:00.847 --> 00:36:04.427
will fire relative to these, indeed, looking like object-based cells.

00:36:04.427 --> 00:36:08.747
But when they placed three wine bottles in a row together, they started to have more influence.

00:36:09.147 --> 00:36:12.187
Probably because of an extended thing, they would be driving more boundary vector

00:36:12.187 --> 00:36:15.847
cells. But the same kind of experiment hasn't been done with these object vector cells yet.

00:36:16.207 --> 00:36:18.287
But the interesting thing about

00:36:18.287 --> 00:36:24.107
now these object vector cells is that they appear in CA1, apparently.

00:36:24.467 --> 00:36:27.347
Well, a very small proportion of CA1 cells, yes.

00:36:27.347 --> 00:36:33.787
Yes, but Edvard was reporting a medial entorhinal and some interesting sort

00:36:33.787 --> 00:36:40.227
of memory object, well, memory times object cells had been reported in lateral entorhinal.

00:36:40.727 --> 00:36:46.027
But can't we, but these are highly plastic circuits and they are exposed to

00:36:46.027 --> 00:36:50.087
multiple, let's say streams of modalities and submodalities.

00:36:50.787 --> 00:36:56.147
So if we just combine these two considerations, is it then not sort of predictable

00:36:56.147 --> 00:37:02.627
that these kinds of combinatorial or conjunctive encodings emerge rather naturally.

00:37:04.047 --> 00:37:15.587
Well, yes, but before we proposed them from our analysis of play cell firing, nobody knew that.

00:37:16.715 --> 00:37:19.495
An offset vector and essentially offset vector was one of

00:37:19.495 --> 00:37:22.575
the ingredients that would get combined with boundary detectors or

00:37:22.575 --> 00:37:25.615
whatever and until the moses recorded grid cells

00:37:25.615 --> 00:37:30.395
nobody thought there's this sort of furrier like representation of space would

00:37:30.395 --> 00:37:35.715
would exist or needed to exist and so yes in retrospect but before the fact

00:37:35.715 --> 00:37:40.515
no no i didn't want to trivialize it i just want to say this might mean that

00:37:40.515 --> 00:37:45.595
if you now go for some ambiguous border object object and for sure

00:37:45.675 --> 00:37:48.575
you'll find a cell that'll respond to it because we look at this combinatorics.

00:37:49.135 --> 00:37:53.415
Maybe, but I mean, the interesting thing, I think, as Edvard pointed out about

00:37:53.415 --> 00:37:58.495
the hippocampus and entorhinal cortex is that coming from the point of view

00:37:58.495 --> 00:38:00.455
of a Hopfield model for memory,

00:38:00.635 --> 00:38:05.275
you might expect a random distributed binary code that was uninterpretable.

00:38:05.335 --> 00:38:09.315
And in fact, you see these incredibly clear, discrete responses,

00:38:09.575 --> 00:38:17.315
say to compass direction or to location, or a Fourier-like like grid or,

00:38:17.475 --> 00:38:19.355
you know, a boundary vector.

00:38:19.495 --> 00:38:22.915
Now there are, as Edvard said, you know, you do get conjunctive directional

00:38:22.915 --> 00:38:25.855
grid cells, for example, and there may be other conjunctions,

00:38:25.895 --> 00:38:30.115
but the overall impression is that, and there are plenty of spatially modulated

00:38:30.115 --> 00:38:34.275
cells that haven't been characterized also in entorhinal cortex.

00:38:35.995 --> 00:38:40.075
They're probably now less than, you know, they're probably a minority of all

00:38:40.075 --> 00:38:44.635
the cells because so many cells have been characterized. But still,

00:38:44.675 --> 00:38:50.355
the overriding impression is of these incredibly discrete types of encoding being present.

00:38:50.495 --> 00:38:54.455
And sure, they are present in some combinations also, but overall,

00:38:54.515 --> 00:38:56.255
you wouldn't say this is just a mush of everything.

00:38:56.535 --> 00:39:00.375
You would say, Jesus Christ, why are those such specific responses there?

00:39:00.375 --> 00:39:05.855
But as a mathematician, it must appeal to you to perhaps to find some theory

00:39:05.855 --> 00:39:09.335
that explains the emergence of these different kinds of cell types,

00:39:09.575 --> 00:39:12.275
which are particularly useful for navigation.

00:39:12.595 --> 00:39:17.455
And then to be able to perhaps generalize that and say what further kinds of

00:39:17.455 --> 00:39:22.255
cell types we might be able to predict that we should see, perhaps because they're

00:39:22.255 --> 00:39:25.875
adding more orthogonal information to the mix,

00:39:26.035 --> 00:39:28.415
perhaps sensory cells, for example.

00:39:29.015 --> 00:39:32.735
And I think one of the things you alluded to, and also Edvard in his talk,

00:39:32.835 --> 00:39:38.415
was how the grid cells have this sort of change in scale, which follows a sort

00:39:38.415 --> 00:39:41.955
of mathematical law, which is consistent with efficient coarse coding.

00:39:42.115 --> 00:39:48.255
So it looks as though there is, again, something which doesn't look at all accidental,

00:39:48.375 --> 00:39:55.135
but must be the result of this system responding in some very effective way

00:39:55.135 --> 00:39:58.275
to choose the right inputs for mapping space.

00:39:59.503 --> 00:40:04.243
Yes, I think when you come to the grid cells and the discrete jumps in scale.

00:40:05.503 --> 00:40:11.383
That is clearly well arranged for large-scale navigation.

00:40:11.863 --> 00:40:18.423
You get a big combinatorial power for the range over which you can encode locations

00:40:18.423 --> 00:40:19.903
with that kind of representation.

00:40:20.683 --> 00:40:23.263
So what is your prediction for the next kind of cell type?

00:40:25.563 --> 00:40:26.123
Um...

00:40:29.843 --> 00:40:34.963
And it's funny, I don't think the field at the moment is working that way.

00:40:35.063 --> 00:40:39.543
It's true that we predicted boundary vector cells from looking at how place cells responded.

00:40:39.663 --> 00:40:45.323
And going back, that argument, that way of thinking was certainly used by O'Keefe and Nadel.

00:40:45.623 --> 00:40:49.143
And John O'Keefe, having found place cells, predicted there should be directional

00:40:49.143 --> 00:40:52.803
cells and something like path integration or distance cells.

00:40:53.163 --> 00:40:58.443
And so early on, having found place cells, he did predict the head direction

00:40:58.443 --> 00:41:00.463
cells and then something to do with distance.

00:41:00.843 --> 00:41:04.263
And they have been borne out. And something to do with vectors,

00:41:04.303 --> 00:41:08.063
it seems like, is necessary. And we have boundary vector cells and object vector cells now.

00:41:09.543 --> 00:41:12.503
I think the interesting thing, I mean, so there's already a lot of ingredients

00:41:12.503 --> 00:41:15.223
there for making quite a sophisticated system for space.

00:41:15.483 --> 00:41:19.783
And at the moment, I think people are thinking more, how could all this stuff

00:41:19.783 --> 00:41:24.883
we've learned about space tell us about more general memory properties of the hippocampal system?

00:41:24.883 --> 00:41:28.183
That you know how does it store other information which

00:41:28.183 --> 00:41:30.943
which we know that the human hippocampus is required for all

00:41:30.943 --> 00:41:34.383
sorts of forms of memory and i think that's interesting

00:41:34.383 --> 00:41:37.183
so i haven't predicted any more kinds of

00:41:37.183 --> 00:41:40.123
spatial cells i must say but that's particularly what i was getting at you know

00:41:40.123 --> 00:41:45.623
sort of human episodic memory which is more than about a lot of things more

00:41:45.623 --> 00:41:52.823
than space um and where it's it's less easy sort of coming at it from the geometry

00:41:52.823 --> 00:41:56.823
or whatever to say what kind of basis functions you would want to have.

00:41:56.943 --> 00:42:02.163
So we might look at some mathematical way of finding appropriate basis functions

00:42:02.163 --> 00:42:04.703
for learning about episodes in time.

00:42:04.803 --> 00:42:10.983
Yes. Well, so I think it's very interesting that the cells in the human hippocampus,

00:42:10.983 --> 00:42:13.803
most famously Jennifer Aniston type cells, are.

00:42:15.032 --> 00:42:20.832
They're a local tuning curve to a concept, in this case of Jennifer Aniston,

00:42:20.932 --> 00:42:23.992
among all famous people or people you might have heard of.

00:42:24.172 --> 00:42:30.192
And a cell will fire to that concept, whether it's a written name or however you bring it to mind.

00:42:30.432 --> 00:42:37.772
And other cells will fire for other concepts, be they spiders or semantic well-known

00:42:37.772 --> 00:42:39.552
landmarks or famous actors.

00:42:39.552 --> 00:42:43.192
And so you might indeed see

00:42:43.192 --> 00:42:46.192
the sort of locally tuned place cell response as very

00:42:46.192 --> 00:42:50.052
similar to these locally tuned responses in semantic space and

00:42:50.052 --> 00:42:53.232
this very interesting um stretchy bird experiment by

00:42:53.232 --> 00:42:58.012
tim behrens and his group showing what looked like grid-like responses in in

00:42:58.012 --> 00:43:05.192
humans to semantics two-dimensional spaces and uh in rats to one-dimensional

00:43:05.192 --> 00:43:10.692
continuously varying tones showing play cell-like responses from David Tank's group,

00:43:11.692 --> 00:43:19.552
do imply that these principles that have been easiest to spot in space and freely navigating animals.

00:43:21.073 --> 00:43:24.393
Might well apply to all sorts of other kinds of information and how it's organized

00:43:24.393 --> 00:43:28.593
and how it's retrieved and so on. And then that would obviously be a...

00:43:29.253 --> 00:43:31.733
Starting with these very well-characterized spatial responses,

00:43:31.833 --> 00:43:35.393
I think, is a good place to try and understand everything else,

00:43:35.433 --> 00:43:39.113
which is obviously much harder to get a grip on.

00:43:39.473 --> 00:43:43.753
So, I mean, that would suggest that one thing we might look for is sort of compression-type

00:43:43.753 --> 00:43:46.893
algorithms, which will give us a low-dimensional.

00:43:47.653 --> 00:43:51.913
Or relatively low-dimensional description of the current context,

00:43:52.193 --> 00:43:59.113
which we can encode then in entron and cortex and use to run our sort of location

00:43:59.113 --> 00:44:01.453
finder or memory finder algorithm.

00:44:01.913 --> 00:44:08.353
Is that a good metaphor? Well, the grid cells certainly appear to be a compressed

00:44:08.353 --> 00:44:12.993
code like a Fourier code or something that would be very efficient compared to the place cells,

00:44:13.153 --> 00:44:18.013
which in their own way are very useful for tagging specific information to specific places,

00:44:18.233 --> 00:44:25.713
but are less efficient for covering large areas in a metric way where you can

00:44:25.713 --> 00:44:28.073
infer the vector between any two parts.

00:44:28.213 --> 00:44:31.413
If I have a place cell that fires in one place and another one that fires in

00:44:31.413 --> 00:44:35.653
another place with no overlap, it's hard to know how they're related to each other.

00:44:35.753 --> 00:44:42.433
Whereas with grid cells, you can infer the vector between two grid cell codes

00:44:42.433 --> 00:44:45.753
for two different locations. And so that's very powerful. It's a compressed representation.

00:44:46.493 --> 00:44:49.293
You know, it looks a bit like a Fourier pattern. But equally,

00:44:49.553 --> 00:44:54.633
Dory Durdickman's group has pointed out that if you do PCA on the place cell

00:44:54.633 --> 00:45:01.233
input that you get as your animal is running around, then you get things that look like grid cells.

00:45:01.233 --> 00:45:05.213
So indeed, the grid cells might be doing a PCA of this place representation.

00:45:06.413 --> 00:45:12.513
And there's a similar argument to be made for the successor representation that

00:45:12.513 --> 00:45:17.093
I mentioned for place cells in that the eigenvectors of the successor representation

00:45:17.093 --> 00:45:18.793
also look like grid cells.

00:45:19.013 --> 00:45:21.573
And so they're clearly related, the.

00:45:22.693 --> 00:45:25.373
Eigenvectors of the covariance matrix being the same as the

00:45:25.373 --> 00:45:28.293
pca it could well be that the grid cells are a

00:45:28.293 --> 00:45:31.593
compressed way of arbitrarily representing

00:45:31.593 --> 00:45:37.933
any of this location coded or you know single peaked tuning curve uh grandmother

00:45:37.933 --> 00:45:42.813
type cell response uh representation of arbitrary information which is very

00:45:42.813 --> 00:45:47.553
useful for then associating things to that information in the hippocampus yeah

00:45:47.553 --> 00:45:51.533
so i think if i maybe put words into your mouth but you're agreeing that

00:45:51.933 --> 00:45:55.253
the entorhinal is a compressed representation of the context,

00:45:55.433 --> 00:45:59.733
and then that's expanded out in CA1, CA3,

00:46:00.353 --> 00:46:04.733
perhaps after being sparsified in dentate gyrus, as often as that people have.

00:46:04.773 --> 00:46:13.673
Well, you know, David Marr's model of hippocampus remains the best model of hippocampus in memory.

00:46:14.273 --> 00:46:18.013
You know, it's probably surviving that and his cerebellum model and surviving

00:46:18.013 --> 00:46:21.173
really well, maybe even better than his later work in vision.

00:46:21.533 --> 00:46:27.353
Um and yes those ideas hold a lot of currency although it's important to remember

00:46:27.353 --> 00:46:32.173
entorhinal cortex is not just grid cells so even within medial entorhinal cortex

00:46:32.173 --> 00:46:36.113
it may be that head direction cells are more numerous than grid cells and there

00:46:36.113 --> 00:46:37.313
are also these border cells.

00:46:38.193 --> 00:46:45.113
Conjunctive cells and spatially modulated but not grid cell cells so you know

00:46:45.113 --> 00:46:48.213
yes for the for the grid cells they look like compressed code uh there's other

00:46:48.213 --> 00:46:51.693
things going on also uh sensory inputs to place cells,

00:46:51.833 --> 00:46:54.213
we think, not just the path integration type inputs.

00:46:54.233 --> 00:46:57.813
But aren't you, first you guys are all swept away by Bayesian,

00:46:57.933 --> 00:47:02.353
and now you're all swept away by, let's say, compression, which in some sense

00:47:02.353 --> 00:47:05.113
is already happening to a large extent outside of Hippocampus.

00:47:05.253 --> 00:47:09.273
And earlier, I thought we had agreed that Hippocampus is contributing to constructing

00:47:09.273 --> 00:47:10.233
allocentric representations.

00:47:11.193 --> 00:47:15.973
So, let's not worry about compression then, but now you guys are all worried about compression.

00:47:16.213 --> 00:47:19.013
So, what is it? Compression, allocentric, or is

00:47:19.013 --> 00:47:23.373
it both what do you want uh well you're

00:47:23.373 --> 00:47:27.053
comparing apples and oranges i i don't i think you

00:47:27.053 --> 00:47:29.893
can have an allocentric or an egocentric representation and you

00:47:29.893 --> 00:47:32.573
might want it to be compressed or not so we i don't

00:47:32.573 --> 00:47:36.253
know we can compare allocentric egocentric no no you have no no what i'm saying

00:47:36.253 --> 00:47:40.413
compression or allocentric is it both is it combined is one building on the

00:47:40.413 --> 00:47:45.313
other are these independent orthogonal interpretations well so so the those

00:47:45.313 --> 00:47:49.393
codes the head direction cells and the place cells and grid cells appear to

00:47:49.393 --> 00:47:51.733
be allocentric in the sense that...

00:47:53.391 --> 00:47:56.551
The reference to the world, it's which way you're facing in the world,

00:47:56.591 --> 00:47:57.431
where you are in the world.

00:47:57.631 --> 00:48:05.091
But they are where the animal currently is, which you might say is egocentric.

00:48:05.151 --> 00:48:07.151
It's referring to the animal's own location or orientation.

00:48:07.811 --> 00:48:13.091
Having said that, I think then the grid cells look like a compressed version of the play cells.

00:48:13.731 --> 00:48:17.891
So it may be that an expanded representation is useful for attaching things

00:48:17.891 --> 00:48:19.611
to, and that might be the play cells.

00:48:19.951 --> 00:48:26.591
And a compressed representation might be useful for generating large-scale metrics

00:48:26.591 --> 00:48:31.751
within which to compare or know the relative positions of lots of information.

00:48:32.151 --> 00:48:36.231
About 10 years ago, we showed that grid cells actually carry more information

00:48:36.231 --> 00:48:37.471
about space than place cells.

00:48:37.751 --> 00:48:40.491
So if you do position reconstruction from grid cells versus place cells,

00:48:40.651 --> 00:48:42.611
you have more accuracy from the grid cells.

00:48:42.831 --> 00:48:45.291
So that would be then inconsistent with the idea that they compress the place

00:48:45.291 --> 00:48:48.631
cells because then the place cells would have more information for position

00:48:48.631 --> 00:48:52.051
reconstruction. Are you comparing the same number of place cells and grid cells?

00:48:52.231 --> 00:48:53.251
Yeah, this was all control.

00:48:53.531 --> 00:48:56.591
Yeah, no, well, exactly. So if you have a sparse representation,

00:48:56.671 --> 00:49:00.831
which is easy to attach things to, then obviously it's more efficient. It is compressed.

00:49:01.051 --> 00:49:06.451
You can have a coverage of a larger environment with the same number of grid

00:49:06.451 --> 00:49:08.731
cells. That's why it's a compressed representation.

00:49:10.191 --> 00:49:18.711
Okay, look. But now the other... So this is a controversy we saw,

00:49:18.711 --> 00:49:19.771
which we will solve later,

00:49:19.971 --> 00:49:24.891
but now you presented a model where you wanted to explain how now you could

00:49:24.891 --> 00:49:31.031
use allocentric representations to also reconstruct, if you want, imagined locations,

00:49:31.311 --> 00:49:37.771
how you could use imagery to sort of make now predictions that would be relevant for navigations.

00:49:38.391 --> 00:49:42.471
What's the contribution of that model now to our understanding of the hippocampus?

00:49:44.623 --> 00:49:49.123
Well, that model was trying to take these sort of spatial things that we recorded

00:49:49.123 --> 00:49:53.203
in and around the campus and say, how could they apply to human episodic memory?

00:49:53.523 --> 00:50:00.263
And, you know, the experience of human episodic memory is reliving the event.

00:50:00.543 --> 00:50:05.463
And in terms of the visual content, it's imagining the scene.

00:50:05.463 --> 00:50:10.583
And so the proposal was that this spatial system is a way of pulling out the

00:50:10.583 --> 00:50:14.823
stored allocentric or abstract information that you have in your temporal lobes

00:50:14.823 --> 00:50:20.963
to create a coherent spatial scene from a single location with a single direction,

00:50:21.263 --> 00:50:24.903
you know, provided by the head direction cells and the place cells respectively.

00:50:26.283 --> 00:50:30.763
And then how you could project that information into an egocentric frame so that you can imagine it.

00:50:30.763 --> 00:50:33.423
So that's a necessary thing that has to happen if we want

00:50:33.423 --> 00:50:36.503
to make contact with with human imagery and so

00:50:36.503 --> 00:50:43.643
that's why i put it in the model um in principle you could solve vector navigation

00:50:43.643 --> 00:50:50.003
just using the the grid and place cell model and maybe do that you need this

00:50:50.003 --> 00:50:54.783
extra stuff to be able to sort of visualize uh but of course that gives you

00:50:54.783 --> 00:50:56.163
know extra functionality you might Right.

00:50:56.963 --> 00:51:01.063
In principle, if you've got a lot of stored abstract information,

00:51:01.383 --> 00:51:09.283
you could pull it together to give you the viewpoint, if you like,

00:51:09.423 --> 00:51:13.703
on all of that information that's consistent with being in a single location.

00:51:14.023 --> 00:51:19.903
And that's a very powerful way of reducing the enormous amount of stored information.

00:51:20.043 --> 00:51:23.063
You're just going to pull out all those bits that are consistent with me being

00:51:23.063 --> 00:51:25.343
here, possibly facing this direction.

00:51:25.343 --> 00:51:28.083
Now what what is around me and if

00:51:28.083 --> 00:51:31.443
you think conceptually you've got

00:51:31.443 --> 00:51:34.223
a whole lot of stored information a whole sort of life's worth of experience

00:51:34.223 --> 00:51:37.143
if you like uh if i ask you about jennifer

00:51:37.143 --> 00:51:40.783
aniston you probably want to pull out just the information relative to

00:51:40.783 --> 00:51:43.983
to that location in in the conceptual space

00:51:43.983 --> 00:51:47.883
of all actors or people that you've ever seen and so

00:51:47.883 --> 00:51:52.383
uh you know it's more that these are

00:51:52.383 --> 00:51:55.483
ways in which memory could function you have to consider the retrieval

00:51:55.483 --> 00:51:58.383
process you could store an awful lot of information but

00:51:58.383 --> 00:52:02.203
how do you retrieve it and episodic memory seems to a very specific way of retrieving

00:52:02.203 --> 00:52:08.983
information you impose a particular location and perhaps a particular direction

00:52:08.983 --> 00:52:12.323
to interrogate your stored data

00:52:12.323 --> 00:52:17.283
with but now in the model itself gives you like a lookup table, right?

00:52:17.343 --> 00:52:20.523
You can sort of project back from the egocentric views to the allocentric views.

00:52:21.683 --> 00:52:25.243
And to what extent is the model then able to capture also the physiological

00:52:25.243 --> 00:52:29.623
and anatomical characteristics of this hippocampal system? Yeah.

00:52:31.390 --> 00:52:35.390
Well, it's quite a high-level model because it's trying to deal with cognition,

00:52:35.490 --> 00:52:42.790
and we know so little that it would be inappropriate to try and build too much

00:52:42.790 --> 00:52:45.470
physiological and anatomical detail in from the start.

00:52:45.730 --> 00:52:51.050
But this translation circuit, as shown by Alex Puget and various other people,

00:52:51.170 --> 00:52:56.290
corresponds to the observed parietal gain field neurons that have been recorded

00:52:56.290 --> 00:53:04.170
and should function as a translation circuit irrespective of what it is you're retrieving,

00:53:05.010 --> 00:53:08.490
from allocentric or egocentric coordinates and translating into the other.

00:53:09.790 --> 00:53:13.430
I'm not sure if that answers your question, actually. Well, because it's also

00:53:13.430 --> 00:53:17.770
a bit a step towards the experiments you did with human subjects where you then

00:53:17.770 --> 00:53:23.870
start to match the response of the model to fMRI experiments with human subjects

00:53:23.870 --> 00:53:26.010
that were in some sort of navigational test.

00:53:26.010 --> 00:53:30.530
So how well was the match then between the fMRI results and the predictions of your model?

00:53:31.310 --> 00:53:39.270
Well, in terms of the fMRI, it was more that we could have, to most of the areas

00:53:39.270 --> 00:53:43.510
of activity, the areas in the brain that showed increased metabolic activity

00:53:43.510 --> 00:53:44.850
during this kind of task,

00:53:45.130 --> 00:53:48.370
we could ascribe some kind of putative function.

00:53:48.630 --> 00:53:52.670
It's very speculative, but the model says what that activity should be actually

00:53:52.670 --> 00:53:55.650
doing. you know in the hippocampus the play style should be doing this and in

00:53:55.650 --> 00:54:00.050
retrospinal cortex it should be a translation circuit or whatever and so it's

00:54:00.050 --> 00:54:01.430
not really that there was a um.

00:54:03.123 --> 00:54:06.023
Particular the model wasn't good enough

00:54:06.023 --> 00:54:11.023
to constrain um what activity

00:54:11.023 --> 00:54:13.963
you should see it's more that my choices of

00:54:13.963 --> 00:54:17.103
where these bits of the model should be in the brain were broadly

00:54:17.103 --> 00:54:20.763
borne out and they give us some kind of way of trying to understand what that

00:54:20.763 --> 00:54:24.983
activity might represent in terms of mechanism so the model would be as heuristic

00:54:24.983 --> 00:54:28.463
to help you not to further think it helped to interpret those results it wasn't

00:54:28.463 --> 00:54:33.083
that we predicted results and you know We could measure the error between the

00:54:33.083 --> 00:54:34.923
predicted results and the actual results.

00:54:35.283 --> 00:54:41.443
But now, in terms of the performance, so you had humans in a virtual reality

00:54:41.443 --> 00:54:45.043
environment that they navigated around.

00:54:45.183 --> 00:54:47.383
There was a target that they had to sort of memorize.

00:54:47.943 --> 00:54:53.103
So we did do a behavioral experiment where we did match the behavioral predictions

00:54:53.103 --> 00:54:54.623
of the model and the behavior of the people.

00:54:55.263 --> 00:54:59.303
And in that case, yes, of the models we tried,

00:54:59.303 --> 00:55:04.583
the boundary vector cell place cell model was the best match to the distribution

00:55:04.583 --> 00:55:09.123
of responses across participants of where they thought this thing was when you

00:55:09.123 --> 00:55:11.383
change the shape and size of the virtual room.

00:55:11.723 --> 00:55:17.603
Right. But we discussed this also during your talk which is I think an interesting

00:55:17.603 --> 00:55:22.003
challenge now because you basically took the response of all the subjects.

00:55:22.123 --> 00:55:26.383
The subjects had to estimate where they were relative to this target.

00:55:26.643 --> 00:55:30.783
So this gives you distribution over all estimates over all subjects, right?

00:55:31.083 --> 00:55:32.683
And then you show that the model,

00:55:32.763 --> 00:55:36.923
your model could capture well the distribution over that whole group.

00:55:38.070 --> 00:55:41.990
But now if we split it out to the individual level, and also you showed that

00:55:41.990 --> 00:55:47.470
data, you actually see that the individual distributions can have a rather different shape.

00:55:47.530 --> 00:55:50.870
Actually, it's clustered in, let's say, three different groups.

00:55:50.910 --> 00:55:53.830
Much less variance within subject than between subject.

00:55:54.210 --> 00:55:58.770
So then in the end, what are we modeling, right? Are we modeling statistically

00:55:58.770 --> 00:56:06.190
over populations, or do we have to take into account, must we be able to also

00:56:06.190 --> 00:56:07.890
account for these individual differences?

00:56:08.070 --> 00:56:13.650
Because you saw that the subjects that were accurate had a pretty nicely clustered set of responses.

00:56:14.230 --> 00:56:18.730
And indeed, the ones who were inaccurate. Well, to be fair, actually,

00:56:18.870 --> 00:56:22.970
in that experiment, because you've changed the environment, there's no right or wrong, in fact.

00:56:23.250 --> 00:56:28.430
You know, you can say... They had a task. They had a task to put their marker

00:56:28.430 --> 00:56:29.930
where they thought the flag was.

00:56:30.070 --> 00:56:32.650
That's it. But the encoding environment had changed.

00:56:32.810 --> 00:56:35.850
Sure. So it's a bit of an arbitrary question, and they did their best.

00:56:35.930 --> 00:56:37.810
And some people, they made different guesses.

00:56:38.070 --> 00:56:40.030
But to answer your question about what the model was showing,

00:56:40.110 --> 00:56:43.350
I think the model was showing that as a population,

00:56:44.110 --> 00:56:49.770
people had a vague idea where things were that related to the walls of the environment,

00:56:49.850 --> 00:56:55.370
and the model captured some aspect of that but didn't capture the individual

00:56:55.370 --> 00:56:56.890
differences by any means at all.

00:56:57.690 --> 00:57:03.430
But it was interesting that as a group, they kept within this sort of very broad framework.

00:57:06.149 --> 00:57:08.829
Distribution that the the model could show i mean we could have

00:57:08.829 --> 00:57:11.769
cranked up the model and said just sliced off the very top uh

00:57:11.769 --> 00:57:15.489
likelihood match for location and

00:57:15.489 --> 00:57:19.089
that would have been you know a very narrow uh distribution

00:57:19.089 --> 00:57:23.489
and some subjects put their responses there and others put them miles away and

00:57:23.489 --> 00:57:27.209
some subjects were also confused right did unacceptable things but looking at

00:57:27.209 --> 00:57:30.309
the at the distribution across the subjects we see three groups right there's

00:57:30.309 --> 00:57:34.189
one group that say let's call them accurate and they're they're at the northeast

00:57:34.189 --> 00:57:36.149
corner of the environment close to the wall.

00:57:36.549 --> 00:57:40.849
The second cluster is sort of southeast, so close to the wall,

00:57:40.909 --> 00:57:43.989
but they're like reversed, right? They've mirrored, if you want that space.

00:57:44.309 --> 00:57:46.989
And the third group is completely off in the opposite corner.

00:57:47.249 --> 00:57:52.109
So is there a set of parameters in your model that would be able to then tune

00:57:52.109 --> 00:57:55.309
the model to capture each of those clusters of responses accurately?

00:57:55.949 --> 00:58:00.869
Well, I think that there was one participant that was off in the corner.

00:58:01.009 --> 00:58:04.369
I think we have to just ignore, who knows what

00:58:04.409 --> 00:58:07.209
they were thinking about then there's two other groups of

00:58:07.209 --> 00:58:10.089
responses and indeed they're not uh one is within

00:58:10.089 --> 00:58:13.049
to see the blue one there is actually some they've responded in

00:58:13.049 --> 00:58:16.089
both um groups um but

00:58:16.089 --> 00:58:18.869
it seems like of these subjects more of them

00:58:18.869 --> 00:58:21.749
because the original location was nearer to

00:58:21.749 --> 00:58:24.789
the north wall and the south wall more of them were if you

00:58:24.789 --> 00:58:27.609
like trying to replicate that distance to the south wall although

00:58:27.609 --> 00:58:31.249
they don't replicate that actual distance it's stretched down somewhat and

00:58:31.249 --> 00:58:34.249
then fewer of them are perhaps if you like paying more

00:58:34.249 --> 00:58:37.649
attention during encoding to the distance to the south wall

00:58:37.649 --> 00:58:40.809
and they're replicating that to a greater extent and so

00:58:40.809 --> 00:58:46.069
who knows as I showed their viewing direction it's not that the ones paying

00:58:46.069 --> 00:58:48.889
attention to the north wall were all looking at the north wall some of them

00:58:48.889 --> 00:58:52.409
were but there's a few that were looking at the north wall that matched the

00:58:52.409 --> 00:58:56.749
distance from the south sure but it's interesting because then the model replicated

00:58:56.749 --> 00:58:59.529
something that was not present in any of your subjects Dudes.

00:59:00.801 --> 00:59:03.981
Well, we did discuss this during the talk. I think some of these subjects who

00:59:03.981 --> 00:59:06.441
are what you call quite accurate. The blue one, okay.

00:59:06.581 --> 00:59:09.601
Yeah, well, and maybe this sort of reddish one that's in there as well.

00:59:09.661 --> 00:59:11.401
A few of them may have actually fitted.

00:59:11.601 --> 00:59:14.841
But yes, the variance across subjects is not captured in the model,

00:59:14.881 --> 00:59:19.001
and it seems like you'd have to have different weights for the north-boundary

00:59:19.001 --> 00:59:22.281
vector cell and the south-boundary vector cell and tweak that between subjects

00:59:22.281 --> 00:59:25.361
to get this sort of response in a way which wouldn't be very satisfactory.

00:59:25.781 --> 00:59:31.181
No, exactly. But you could do many, many trials and then see if you're still

00:59:31.181 --> 00:59:36.361
predicting their responses, if you could find some way of knowing which wall

00:59:36.361 --> 00:59:38.741
they're going to be paying a bigger weight to.

00:59:38.941 --> 00:59:44.221
We didn't try and do that, but you might be able to. But do you see this as

00:59:44.221 --> 00:59:48.241
a critical challenge to the model, or you think that this is a detail that you can handle?

00:59:50.681 --> 00:59:53.881
Well, so that model only

00:59:53.881 --> 00:59:56.861
requires the boundary only vector cells and the place cells and so now

00:59:56.861 --> 00:59:59.881
we have this more elaborated model which includes the the visual visual

00:59:59.881 --> 01:00:03.221
imagery and would allow us to also incorporate what some of them seem to be

01:00:03.221 --> 01:00:07.981
trying to do by lining up their viewpoint at encoding and retrieval which would

01:00:07.981 --> 01:00:12.661
be to match the current visual scene with the imagined visual scene from encoding

01:00:12.661 --> 01:00:18.781
we could try to add that that would be you know these three different frameworks that i mentioned at the

01:00:18.801 --> 01:00:22.401
beginning i think the spatial

01:00:22.401 --> 01:00:25.381
updating or path integration probably can't be

01:00:25.381 --> 01:00:28.441
used here because there's a random teleportation to

01:00:28.441 --> 01:00:32.581
a different spot uh before they make their response but the other two the visual

01:00:32.581 --> 01:00:38.801
matching and the environmental location surely are relevant so we could try

01:00:38.801 --> 01:00:42.601
adding the visual matching to this and see if that improves and maybe if something

01:00:42.601 --> 01:00:46.981
about the the orientation at encoding gives some of this inter-subject variation,

01:00:47.601 --> 01:00:49.701
because the encoding positions are all different.

01:00:49.981 --> 01:00:53.241
That might be a way to go. We haven't tried to do that so far.

01:00:53.421 --> 01:00:57.121
But that would mean that models on navigation would have to start to include

01:00:57.121 --> 01:00:58.481
a notion of individual style.

01:01:00.621 --> 01:01:04.261
Only if you want. I mean, yes, if you're interested in those inter-individual

01:01:04.261 --> 01:01:08.201
differences, or you might say like we did in this original experiment, who cares?

01:01:08.241 --> 01:01:11.381
Let's see if we can get any kind of match to anything averaged over whatever.

01:01:11.381 --> 01:01:15.761
Yeah, but now you have matched to an average person that doesn't exist. Yeah.

01:01:18.041 --> 01:01:22.041
It's a typical problem in modeling, right? Well, and other good points could

01:01:22.041 --> 01:01:24.281
be made, like all models are wrong. Yeah, sure.

01:01:24.981 --> 01:01:28.121
But actually, a model is only replaced by a better model also.

01:01:28.661 --> 01:01:33.461
Yeah, no, no, absolutely right. So, look…,

01:01:35.730 --> 01:01:43.670
Let's get to the – you then started to map these ideas, also reasoning from these models.

01:01:43.750 --> 01:01:47.470
You looked at the response in the human brain using fMRI.

01:01:47.750 --> 01:01:51.010
But now the second problem that you could face there is, of course,

01:01:51.050 --> 01:01:52.450
also the matching in time, right?

01:01:52.490 --> 01:01:56.070
Because initially we talked about grid cells, place cells, so on,

01:01:56.150 --> 01:01:59.270
and we talked about electrophysiological data. This is all happening in milliseconds.

01:02:00.570 --> 01:02:05.010
And now we are validating the model against fMRI experiments where we actually

01:02:05.010 --> 01:02:10.350
look at a very different spatial temporal window of response that's measured in seconds, right?

01:02:10.710 --> 01:02:15.370
So how do you overcome that challenge?

01:02:15.490 --> 01:02:18.970
And in some sense, the dynamics of the model is completely outside of the window

01:02:18.970 --> 01:02:22.070
of the measurement technique that you use to validate the model.

01:02:24.490 --> 01:02:31.410
Well, we made predictions that were appropriate for that kind of testing.

01:02:31.410 --> 01:02:36.110
We also use MEG and intracranial recording, which has higher temporal resolution, of course.

01:02:36.390 --> 01:02:42.650
But for predictions for fMRI, the most common kind of prediction is that,

01:02:42.710 --> 01:02:46.230
on average, during this task, this particular part of the brain is going to

01:02:46.230 --> 01:02:49.970
be a bit more active because there's neurons there that are doing something

01:02:49.970 --> 01:02:51.170
that are required for the task.

01:02:51.290 --> 01:02:54.230
And that's the kind of prediction that we made. though you may be referring

01:02:54.230 --> 01:02:58.250
to the sort of grid cell like experiment where we try to make a prediction for

01:02:58.250 --> 01:03:03.970
for what would be the time averaged bold signal as a function of running direction,

01:03:04.630 --> 01:03:11.570
over the whole trial um by noticing that the the grid cell firing patterns tend

01:03:11.570 --> 01:03:16.530
to be aligned across the whole population of grid cells we try to make a prediction

01:03:16.530 --> 01:03:21.710
for a difference according to to alignment with the grid axes or misalignment

01:03:21.710 --> 01:03:23.530
with the grid axes in terms of your running direction,

01:03:24.430 --> 01:03:29.790
which we could look for on this time-average data over the whole trial. Right.

01:03:30.090 --> 01:03:32.270
But that was a brilliant idea.

01:03:32.810 --> 01:03:37.190
How did you stumble into that? Or was it really like a sudden insight,

01:03:37.390 --> 01:03:41.510
like, okay, the orientation of the grids has a certain discretization.

01:03:42.430 --> 01:03:46.090
And given that discretization, we must see alignment or misalignment,

01:03:46.090 --> 01:03:47.830
and we can exploit that in our bolt signal?

01:03:47.990 --> 01:03:52.610
Was it really like... Well, I should say, so the other two authors who were

01:03:52.610 --> 01:03:58.370
both postdocs in my group at the time, Christian Dohler and Caswell Barry, had been wanting...

01:03:58.370 --> 01:04:03.650
So Caswell Barry was recording grid cells in rodents and Christian Dohler was doing fMRI.

01:04:03.790 --> 01:04:09.570
And he really wanted, the two of them really wanted to be able to see something

01:04:09.570 --> 01:04:14.910
in fMRI that might somehow relate to grid cells. And so we did discuss it for,

01:04:14.910 --> 01:04:17.670
you know, some years before.

01:04:18.550 --> 01:04:22.130
And Caswell noticed that these grid cells were aligned, the grid patterns were

01:04:22.130 --> 01:04:25.210
aligned. And so we tried to eventually...

01:04:27.521 --> 01:04:28.861
I had heard of quadrature filters,

01:04:29.041 --> 01:04:31.941
which are a good way of looking for a particular phase orientations.

01:04:32.001 --> 01:04:35.781
And so, you know, we put this method together and eventually applied it to data

01:04:35.781 --> 01:04:41.381
that we'd actually recorded over the previous four or five years for other reasons

01:04:41.381 --> 01:04:46.701
prior to coming up with this particular idea for looking for a grid-like signal.

01:04:47.301 --> 01:04:51.281
Right. It's an amazing result. But now it's the 60 degrees.

01:04:52.701 --> 01:05:00.601
Is that, let's say, the optimal, if you want, misalignment between grid cells?

01:05:00.801 --> 01:05:03.881
Is that where you would find the strongest competition between their responses?

01:05:04.301 --> 01:05:08.761
Or do you see the 60 degrees as some sort of canonical feature of their alignment

01:05:08.761 --> 01:05:10.781
in the human entorhinal cortex?

01:05:10.781 --> 01:05:13.601
No it's it's just that the actual firing

01:05:13.601 --> 01:05:18.141
pattern itself is um a regular

01:05:18.141 --> 01:05:21.381
triangular grid and so it has

01:05:21.381 --> 01:05:24.701
orientational symmetry of 60 degrees and

01:05:24.701 --> 01:05:31.961
so that's why every 60 degrees of rotation you should see a similar if you're

01:05:31.961 --> 01:05:35.561
looking ahead you're running in a straight line ahead then you should see a

01:05:35.561 --> 01:05:40.541
similar pattern of uh firing fields of all the different grid cells in front

01:05:40.541 --> 01:05:43.841
of you because it looks the same every time you go through 60 degrees.

01:05:43.901 --> 01:05:46.501
That's because it's a regular triangular grid.

01:05:46.741 --> 01:05:50.101
But that would mean any multiple of 60 should also work. Yes.

01:05:50.481 --> 01:05:55.641
Yes, it does. I mean, the control experiments in that paper was looking at things

01:05:55.641 --> 01:05:57.121
that were not multiples of 60.

01:05:58.341 --> 01:06:03.361
30 also should not work because that in fact should be the optimal disambiguated.

01:06:03.441 --> 01:06:06.181
And that's what we compared, sort of 30 versus 60.

01:06:06.581 --> 01:06:14.061
But we also looked for you know, instead of six-fold rotational symmetry,

01:06:14.341 --> 01:06:19.081
seven-fold, five-fold, four-fold, eight-fold, and we did not see that.

01:06:19.201 --> 01:06:24.101
So that was the control for our method. Right. So why do you think the triangular...

01:06:25.838 --> 01:06:28.918
Well, actually, that's an interesting point. So if you were an engineer,

01:06:29.118 --> 01:06:31.218
which you probably are, I don't know.

01:06:31.658 --> 01:06:34.818
I'm a psychologist. You're a psychologist, okay. Well, if you were an engineer,

01:06:34.958 --> 01:06:40.578
then you might design some kind of navigation system or something to provide

01:06:40.578 --> 01:06:43.858
a metric on an X and Y right angular axes.

01:06:46.978 --> 01:06:51.878
So why do we have sort of axes that it's more like triangular axes with a 60

01:06:51.878 --> 01:06:52.978
degrees instead of 90 degrees?

01:06:52.978 --> 01:06:57.698
And um one i think um one reason

01:06:57.698 --> 01:07:01.478
you can see from um from dead

01:07:01.478 --> 01:07:07.118
reckoning so in sailors that were navigating they would um try to have multiple

01:07:07.118 --> 01:07:12.898
lines that they were integrating across and uh the reason why three is better

01:07:12.898 --> 01:07:16.658
than two because if i'm estimating my displacement in x and in y it gives me

01:07:16.658 --> 01:07:21.298
a location but if i'm doing it on three axes then any pair of them will give me a location,

01:07:21.418 --> 01:07:24.518
and I can check for my error by comparing it with any other pair.

01:07:24.898 --> 01:07:28.838
And so by having more than two axes, it's redundant, but the redundancy is useful

01:07:28.838 --> 01:07:31.098
because it tells you when you're accumulating error.

01:07:32.298 --> 01:07:38.518
Right. Well, an alternative is that a triangle is the optimal way to cover a

01:07:38.518 --> 01:07:42.738
sphere, which would be interesting for grid cells because that means it's sort

01:07:42.738 --> 01:07:45.358
of a never-ending, right, attractor system.

01:07:45.358 --> 01:07:49.418
Them um well you

01:07:49.418 --> 01:07:52.178
you could so it's close-packed and so

01:07:52.178 --> 01:07:55.918
you might think that you you have um you

01:07:55.918 --> 01:07:59.898
know 2d space represented on a torus in fact yeah and you have close-packed

01:07:59.898 --> 01:08:05.078
representation on that surface and and indeed you know hexagonal close packing

01:08:05.078 --> 01:08:11.298
is is is likely um is another good good reason for why that you might end up

01:08:11.298 --> 01:08:15.818
with grid cells that's right there are models of how how grid cells could be formed, say,

01:08:15.918 --> 01:08:18.458
from place cells in terms of compression or whatever,

01:08:18.618 --> 01:08:21.418
and you would end up with hexagonal closed packing.

01:08:21.618 --> 01:08:26.118
That's right. Although in some cases, if you do the PCA, certainly in rectangle

01:08:26.118 --> 01:08:29.198
environments, you end up with 90-degree grids also.

01:08:31.178 --> 01:08:35.478
So in our models, it's a twisted torus, essentially, with triangular packing

01:08:35.478 --> 01:08:37.138
because that's optimal. Yeah. Okay.

01:08:38.418 --> 01:08:44.338
So the other thing that you showed towards the end of your talk was actually

01:08:44.338 --> 01:08:50.898
an interesting effect that if you have connected rooms but not continuous rooms,

01:08:51.138 --> 01:08:53.838
and you measure the grid cell response,

01:08:54.518 --> 01:09:01.578
along these rooms, that they actually are sort of adjusting themselves with time, right?

01:09:01.898 --> 01:09:08.578
So how large can these adjustments be, right? So what kind of shifts can you observe?

01:09:08.678 --> 01:09:11.418
Because it seemed to look like if these are connected rooms,

01:09:11.418 --> 01:09:17.118
They converge onto still a consistent mapping, a consistent coverage of then

01:09:17.118 --> 01:09:18.778
these two rooms as if it is one.

01:09:21.913 --> 01:09:27.153
So what was the question? Well, what kind of modulation can you expect to see there?

01:09:27.353 --> 01:09:32.253
Well, so Francis Carpenter, who did this experiment with Caswell-Barry,

01:09:32.413 --> 01:09:38.753
he recorded for up to 21 days of experience of the rats walking between these

01:09:38.753 --> 01:09:40.013
two rooms along the corridor.

01:09:40.873 --> 01:09:46.233
And after that long length of time, the grid-like patterns were becoming,

01:09:46.393 --> 01:09:51.233
in the two adjacent boxes, becoming more like a global grid that covered both

01:09:51.233 --> 01:09:53.313
of them. but still hadn't got the whole way there.

01:09:54.313 --> 01:10:00.973
And so I guess what you could expect with experience would be adjustments that

01:10:00.973 --> 01:10:08.573
can cover up to half a grid wavelength so that the two grids can come into alignment

01:10:08.573 --> 01:10:10.493
whatever their starting phases are.

01:10:11.033 --> 01:10:12.913
It's interesting, in that experiment we had.

01:10:15.753 --> 01:10:20.453
Perceptually identical, as far as we could make it, boxes so that the orientations

01:10:20.453 --> 01:10:21.573
of the grids would be aligned.

01:10:21.773 --> 01:10:26.233
I don't know if it might be much harder for the system with misaligned grids

01:10:26.233 --> 01:10:28.293
into two rooms to bring them into alignment.

01:10:28.693 --> 01:10:32.393
Although experiments by Jeff Tauby's group has shown the equivalent thing in head direction cells,

01:10:32.573 --> 01:10:37.953
that if you have two different boxes in which head direction cells have different

01:10:37.953 --> 01:10:41.873
directional tuning and you join them with a corridor that the rat can walk through,

01:10:42.033 --> 01:10:44.733
then eventually they line up and become consistent.

01:10:45.033 --> 01:10:50.113
And I would see these all as part of the same process of trying to build some

01:10:50.113 --> 01:10:54.513
kind of representation of space which is consistent to path integration between

01:10:54.513 --> 01:10:55.953
the different bits of space.

01:10:56.233 --> 01:11:00.173
And do you see that as an adjustment driven by, let's say, CA1?

01:11:01.478 --> 01:11:06.298
Or is it extra Hippocampal? No, I would see in that case,

01:11:06.418 --> 01:11:09.558
because it's the path integration by walking backwards and forwards,

01:11:09.618 --> 01:11:14.518
which is presumably driving the reorganization because that's what tells you

01:11:14.518 --> 01:11:16.798
the relative locations in the two boxes.

01:11:17.178 --> 01:11:22.398
So it's probably the grid cells trying to, if you like, align according to their

01:11:22.398 --> 01:11:24.118
movement-related inputs that

01:11:24.118 --> 01:11:28.718
drives the slow remapping that you see in CA1 in this type of situation,

01:11:28.718 --> 01:11:32.978
where place cells that are originally firing identically in the two boxes,

01:11:33.738 --> 01:11:36.498
begin to slowly disambiguate the two boxes.

01:11:37.018 --> 01:11:42.958
But there's still some monitoring process that can then check whether the grid

01:11:42.958 --> 01:11:48.318
cell response becomes more or less regular at some periodicity, right?

01:11:48.338 --> 01:11:51.578
Otherwise, you won't get the correct alignment because there's a phase and spacing

01:11:51.578 --> 01:11:54.358
and orientation that has to be aligned.

01:11:54.778 --> 01:12:00.758
Yes, I'm not sure there's anything checking. it's more like simultaneous localization and mapping.

01:12:00.878 --> 01:12:04.118
So in robots, you know, you come into a new environment, you try and you've

01:12:04.118 --> 01:12:08.818
got movement detectors, you've got sensory inputs, you try and build a map that's

01:12:08.818 --> 01:12:09.758
consistent with everything.

01:12:10.558 --> 01:12:14.958
And I think in this situation, perceptually identical, the perceptual feed forward,

01:12:15.118 --> 01:12:18.798
you know, boundary vector cells, and so on to place cells was dominating and

01:12:18.798 --> 01:12:21.858
making identical replications of patterns in both boxes.

01:12:22.078 --> 01:12:25.158
But eventually, it manages to do something like slam

01:12:25.158 --> 01:12:27.878
and make a representation which is

01:12:27.878 --> 01:12:31.198
consistent with the path integration okay but

01:12:31.198 --> 01:12:34.018
i would see it as you know you can't put one

01:12:34.018 --> 01:12:36.878
before the other it's trying to come to a compromise of both

01:12:36.878 --> 01:12:39.718
and there's nothing checking okay but the

01:12:39.718 --> 01:12:43.458
plasticity sits in the entorhinal cortex driving

01:12:43.458 --> 01:12:46.458
this or the or in the hip or in the ca1 or ca3 well

01:12:46.458 --> 01:12:50.438
there must be plasticity uh i

01:12:50.438 --> 01:12:54.738
was going yes i'm not sure actually i was what

01:12:54.738 --> 01:12:57.858
i was going to say was there must be plasticity between the play cells and the

01:12:57.858 --> 01:13:02.398
grid cells because the alignment must be but i'm not sure now actually it may

01:13:02.398 --> 01:13:07.218
be that um you know the play cells are trying to anchor the grids to the environment

01:13:07.218 --> 01:13:12.438
and the grid cells providing path integration input to the play cells it may be that.

01:13:13.982 --> 01:13:22.302
That doesn't require plasticity, and that the plasticity happens as the grids, if you like,

01:13:22.342 --> 01:13:26.562
pay more attention to their movement-related inputs than their sensory-driven

01:13:26.562 --> 01:13:29.982
inputs from place cells and become more path integration-driven.

01:13:30.542 --> 01:13:35.702
And that, as a consequence, changes the place cells to become more consistent with that pattern.

01:13:35.902 --> 01:13:39.562
Exactly. And that actually didn't require plasticity between the place cells

01:13:39.562 --> 01:13:40.522
and the grid cells. Right, yeah.

01:13:41.622 --> 01:13:47.062
And on top of that... But on top of that, then you must have plasticity with the two kinds of inputs.

01:13:47.182 --> 01:13:50.982
So I've posited the movement-related input that we think is driving the grid

01:13:50.982 --> 01:13:54.582
cells and the environmental sensory input, which I think is driving the place cells.

01:13:55.322 --> 01:13:59.162
There must be plasticity in both of those inputs, because in the end,

01:13:59.222 --> 01:14:03.322
a place cell in one of the environments will be firing differently relative to its sensory inputs.

01:14:03.682 --> 01:14:09.362
And initially, the grid cells will be firing inconsistently with path integration

01:14:09.362 --> 01:14:10.982
because they're driven by the sensory input.

01:14:10.982 --> 01:14:15.682
So the plasticity in that model would be not between the place cells and grid

01:14:15.682 --> 01:14:18.142
cells, but between the place cells and the central input and the grid cells

01:14:18.142 --> 01:14:19.302
and the path integration. Right.

01:14:19.822 --> 01:14:24.522
But also because this emerges relatively slowly, right? It takes a long time to get this alignment.

01:14:25.002 --> 01:14:28.962
So how many hours does the animal spend between these two? Well,

01:14:28.962 --> 01:14:32.462
that would be a couple of hours a day for many, many days. Exactly.

01:14:32.662 --> 01:14:38.342
And so one valid question is if you made the task depend on knowing where you

01:14:38.342 --> 01:14:42.662
were in each box separately, maybe it would happen faster. Who knows?

01:14:42.842 --> 01:14:46.162
That might be worth trying because it's a horribly long experiment to run in

01:14:46.162 --> 01:14:49.542
its current form. But we don't know. Okay, great.

01:14:50.242 --> 01:14:56.042
So we talked earlier about the concept cells, and in some sense it's now...

01:14:57.050 --> 01:15:01.690
The hippocampus is almost making Jennifer Aniston more famous than she was ever before.

01:15:02.430 --> 01:15:09.630
But we also have a new concept cell, which is now representing the concept cell of Jennifer Aniston.

01:15:10.310 --> 01:15:17.410
So if we start to think about generalizing away from spatial cognition and just behavior in space,

01:15:17.590 --> 01:15:21.430
and we start to think about other domains in which you can use these capabilities

01:15:21.430 --> 01:15:26.850
of the hippocampus, we still would need some sort of path integrator that drives that system.

01:15:27.050 --> 01:15:30.570
So imagine we're going to apply this now to, let's say, a complex cognitive

01:15:30.570 --> 01:15:36.130
task, and we're going to use the computational capabilities of the hippocampus.

01:15:36.190 --> 01:15:42.290
What would then stand in for its path integrator if I'm not moving in physical space anymore?

01:15:42.290 --> 01:15:45.070
More well i think as

01:15:45.070 --> 01:15:48.490
we um discussed with this stretchy bird

01:15:48.490 --> 01:15:51.270
experiment and the grid cells if you have a

01:15:51.270 --> 01:15:56.150
nice grid cell representation it's very powerful and can represent the vector

01:15:56.150 --> 01:16:02.310
2d vector relationships between uh any two points in a very large space this

01:16:02.310 --> 01:16:07.370
could be very useful for mapping all sorts of things concepts for example where

01:16:07.370 --> 01:16:08.870
the vectorial relationship would

01:16:08.910 --> 01:16:14.530
be on vectors of neck length or leg length or who knows how much I like them

01:16:14.530 --> 01:16:18.650
versus how much they wear expensive clothes or anything.

01:16:19.130 --> 01:16:23.490
But it would be a powerful system for representing whatever that conceptual space is.

01:16:23.910 --> 01:16:28.450
And although path integration was probably useful for wiring up these grid cells

01:16:28.450 --> 01:16:32.710
in the spatial situation, if you have that representation somehow.

01:16:33.590 --> 01:16:39.550
In this non-spatial situation, and maybe it's developed as a PCA of these non-spatial firing patterns,

01:16:40.410 --> 01:16:46.170
or maybe it's somehow the conceptual problem is mapped onto space implicitly.

01:16:46.370 --> 01:16:51.330
Either way, you then got a very powerful system for understanding relationships

01:16:51.330 --> 01:16:52.770
between different concepts.

01:16:54.150 --> 01:16:57.450
Okay. So Neil, to finish up,

01:16:58.535 --> 01:17:03.915
okay, you come from physics, went to theory, and then to experiments,

01:17:04.115 --> 01:17:08.255
now you combine this in animals and humans, and you also have been part of some

01:17:08.255 --> 01:17:10.575
really great discoveries in this field,

01:17:11.575 --> 01:17:16.475
and which also was recognized recently by being elected as a fellow of the Royal

01:17:16.475 --> 01:17:20.255
Society, which is a big compliment to your work, so congratulations.

01:17:22.035 --> 01:17:29.955
But the question now is, So given that experience, what would you see as Neil's

01:17:29.955 --> 01:17:31.895
law in the study of the brain?

01:17:33.815 --> 01:17:38.895
I don't think there's any such thing as Neil's law. I do think that if you're

01:17:38.895 --> 01:17:46.375
trying to model something complicated like cognition.

01:17:49.355 --> 01:17:54.335
Then at the level of neurons, you need to start with the simplest possible model

01:17:54.335 --> 01:17:58.815
because we have so little idea about how neurons do represent things like cognition.

01:17:59.215 --> 01:18:04.495
Obviously, in the spatial domain, we have some great clues from all of this work we've heard about.

01:18:05.495 --> 01:18:09.195
And so i tried to always use

01:18:09.195 --> 01:18:13.075
the simplest model um and also

01:18:13.075 --> 01:18:15.955
it might not be a mathematically tractable model although

01:18:15.955 --> 01:18:19.955
it would be great if if it was but it doesn't have to be it's not clear the

01:18:19.955 --> 01:18:24.715
brain is going to be mathematically tractable and it needs to make experimental

01:18:24.715 --> 01:18:31.055
predictions and indeed experiments have to um have something to say to to theory

01:18:31.055 --> 01:18:34.875
other you know they each only exists with the other in some useful sense.

01:18:35.075 --> 01:18:37.235
If you do an experiment, it doesn't impact any theories.

01:18:37.495 --> 01:18:41.715
What was the point? If you have a theory that doesn't impact any experiments,

01:18:41.935 --> 01:18:45.835
you know, maybe it'll be useful one day, but it's not that useful right now.

01:18:45.915 --> 01:18:52.315
But wait, the law of defining laws is they must fit on a t-shirt. So what's the law?

01:18:56.050 --> 01:18:58.850
Uh keep it well i

01:18:58.850 --> 01:19:01.690
think i think there's a um somebody famous said something

01:19:01.690 --> 01:19:04.990
like you know if i could capture my contribution maybe

01:19:04.990 --> 01:19:08.870
it was feinman in in one sentence it wouldn't be much of a contribution okay

01:19:08.870 --> 01:19:12.470
i don't think that's true i think there are some great discoveries which can

01:19:12.470 --> 01:19:15.650
be captured on a t-shirt but i'm not sure but i thought you were saying you

01:19:15.650 --> 01:19:19.210
know i thought you were saying keep it simple yeah that's what i that would

01:19:19.210 --> 01:19:22.170
be fine keep it simple keep it grounded in experiment.

01:19:22.450 --> 01:19:25.190
Okay, good. Great. It's not a very exciting t-shirt.

01:19:25.930 --> 01:19:27.290
Well, it depends what else you put on it.

01:19:28.130 --> 01:19:33.270
Or who's wearing it. Maybe Jennifer Aniston. But look, five years from now,

01:19:33.330 --> 01:19:38.450
I'm going to smuggle myself into the UK because by then, after Brexit and the

01:19:38.450 --> 01:19:42.610
whole disaster that goes along, I will not be allowed to enter anymore in any legal way.

01:19:42.750 --> 01:19:45.610
And I'm going to come to your lab because by then you'll still be there.

01:19:47.390 --> 01:19:53.030
And I'm going to check whether a specific prediction you made today was actually verified or falsified.

01:19:53.310 --> 01:19:58.450
So what's the one prediction you really would like to see tested in that five-year frame?

01:19:59.510 --> 01:20:05.070
Well, there's a few.

01:20:05.230 --> 01:20:08.630
I mean, I didn't really talk about it because there wasn't time and it seemed

01:20:08.630 --> 01:20:11.450
a bit complicated, but I really would like to know if.

01:20:13.490 --> 01:20:17.870
Temporal coding and theta phase procession has something to do with path integration.

01:20:18.510 --> 01:20:25.790
And it may take nearer to 10 years but I hope that one point we'll be able to image grid cells,

01:20:26.470 --> 01:20:30.910
and their dendritic inputs to work out what it is that's making them fire in

01:20:30.910 --> 01:20:35.770
a grid like pattern and there's a clear prediction that you should see oscillations

01:20:35.770 --> 01:20:38.750
of different frequencies in these different dendrites and I would I would like

01:20:38.750 --> 01:20:43.470
to see that at the more cognitive level there's some

01:20:43.510 --> 01:20:48.530
applications to what happens in post-traumatic stress disorder in terms of different

01:20:48.530 --> 01:20:52.850
forms of memory supporting imagery or impacting on them in different ways.

01:20:53.070 --> 01:20:57.030
And it would be nice to see if some of those predictions have come out that

01:20:57.030 --> 01:20:59.670
this model of imagery had some clinical relevance.

01:21:00.470 --> 01:21:04.610
All right, great. Neil Burgess, thank you very much for this conversation. Thank you.

01:21:08.370 --> 01:21:13.950
The CSN Podcast was produced by the Convergent Science Network of Biometrics

01:21:13.950 --> 01:21:20.310
and Biohybrid Systems, a project funded by the European Sevens Research Framework Program.

01:21:21.890 --> 01:21:27.210
For more interviews, recorded lectures, or upcoming conferences in the field

01:21:27.210 --> 01:21:33.450
of biomimetics and biohybrid systems, go to csnnetwork.eu.

01:21:33.680 --> 01:21:42.000
Music.