The famous work of Hubel and Wiesel, mostly done with cats. Sorry, I realize that's going to upset some people. That's just how it was done. Cat is lying on a table looking at a stimulus. There's an electrode in the cat's visual cortex.
And so here's the view looking on at the cat. He's in an apparatus here. There are electrodes there. And out here in front of the cat, Hubel and Wiesel and their colleagues are flashing up light in different shapes and different positions in the cat's visual field.
OK, and recording from neurons. OK, so here is an example. So this is one of Hubel and Wiesel's movies.
This is what the cat is seeing. So they're flashing a bar of light in front of the cat, in which appear are the action potentials of a single neuron in the cat's visual cortex. OK, and so they're marking on a piece of paper what the receptive field of that cell is.
See, they keep moving it around. And once it gets to that edge, it responds. OK, everybody got the idea?
That's how you map out a receptive field. Now there's another movie that I'm running low on time so I think I'll just post it on the Stellar site and you can look at it offline where this goes on for minutes and they change the orientation of that bar and they do all kinds of stuff and it's pretty cool. Okay, but I won't take the time to run through it. The upshot of that is that what you find in primary visual cortex is that neurons have a property called orientation selectivity. And that means that each neuron likes bars of a certain orientation.
That neuron we just saw liked, what was it, like this. Later on in that movie, they present lines like this, bars of light, and it doesn't like those. It likes this, okay? But you say it doesn't like, it means that it doesn't fire as much.
Yeah, yeah. OK. So here is a depiction of that. This is all one neuron. The dotted bar is a receptive field where you have to put stuff to make that neuron fire.
And here's the firing over time when you put stuff in there. What you see is that this neuron responds to bars tilted like this, not bars tilted like that or like that. Everybody see that that's orientation selectivity of that single neuron?
OK. If you plot that, you smoothly vary the orientation. Of that bar in the receptive field and you measure the firing rate of that neuron you get a curve like this also showing orientation selectivity Everybody got that why would a brain why would a visual system have orientation selectivity?
What good is that if you were? Writing up the code to do object recognition. Would you do this at the front end? Why? So one way to look at it is this is sort of the very primitive beginning alphabet out of which we will build shape.
Remember, what the visual system is doing is trying to tell you what's out there. It's got to somehow go from spots of light hitting your retina to Miguel or Stephanie. How's it going to do that? We don't know. But it's going to have to extract information with some kind of building blocks out of which it's going to construct some perceptual representation.
So what we're seeing here is one of the very early stages of building up a perceptual representation. First, finding the bits that matter. in the visual field.
That's what you do with retinal ganglion cells. Find things that change over space and time. Next, start getting primitives of shape.
If you're going to describe the shape of an object, you need to know how the edges are oriented. That's what you see right here. This one fires a lot like that.
It fires a little bit like that and not at all like that. Not at all. Often in the nervous system, it needs background firing rate, the occasional spike that just happens now and then, but much lower firing rate.
OK. That's what this represents. Much more firing to the preferred orientation than the non-preferred orientation.
Now, all of this is sticking electrodes right in, in this case, Cat V1, responding the, recording the firing rate as a function of the orientation of that bar. That's cool. That seems like a sensible way to do it.
Is there any way to detect orientation selectivity just with behavior? It's like, how the hell would you do that? We're looking in the middle of the system. We're recording from neurons.
What can we do with behavior that would tell us about the orientation selectivity or lack thereof in the brain? But there's a way. In fact, this was hypothesized way before Hubel and Wiesel.
Question is, could we discover the same Thing. Can we discover the idea that there are neurons in your visual system tuned to orientations, to specific orientations? Could we discover that without making a measurement from neurons, just measuring behavior? We're going to discover it right now, I hope. It's a slightly weak demo, but I hope it will work.
OK. So now, here's what you need to do. First, look here.
Everybody see nice vertical lines? Got that? Duh. OK, now, your job is to fixate right on that horizontal bar. You can move back and forth along the width of the bar, but you can't leave the bar, and you have to keep fixating.
For a pretty good while, this is a subtle effect. So you're going to have to keep doing this for another 20 seconds or so while I fill in air time. And what you're doing now as you stare at that, hopefully, is tiring out those orientation selective neurons above and below your visual field. Keep fixating there. Keep tiring out those neurons.
And the idea is if you do that long enough, the signal that your brain will be sending up to you, the conscious perceiver, wherever that is, will be a code in which the representation of those orientations has been diminished because you burned them out. You adapted them out. Okay, keep looking for another few seconds.
Don't do this yet, but when I say, what you'll do is you'll shift your gaze over to the horizontal bar. To the left, and it's pretty subtle, but you can tell me if you see anything. OK, try shifting them.
Did it work? Did you see these guys tilted a little bit more like that? Awesome.
OK, this is a tilt after effect. And isn't that cool? Like right here in this class with a projector and a bunch of people, we discovered the properties of neurons in your visual system just by looking at what you see after you stare at this.
OK? Does everybody get the gist of why that would happen? Think of you have these pools of neurons in your primary visual cortex tuned to each of these different orientations. And now what we did was we made you really tire out the neurons that like this, or whatever it was. Yeah.
Look at that long enough. They adapt, just like retinal ganglion cells adapt. OK? Those neurons adapt.
They tire out. They're less interested in firing, just like you run a marathon. You don't want to run anymore.
They're done. Right? And so they are firing less.
And so the net that average orientation indicated by the whole pool of neurons is shifted in the direction of the other ones because they are kind of taken out of your representation. Does that make sense? And it gives you an opposite after effect. Okay.
I mentioned that just to say that it's kind of cheating to record from neurons. It's really, the really hip thing is to infer what the neurons are doing with a nice low tech, I'm sort of kidding, but it's pretty cool to be able to do this without actually recording. The coolest thing actually is having both.
to really make it a strong argument. OK. All right, so this is adaptation is sometimes called the psychophysicist microelectron.
Psychophysicists are people who do just like this. They present visual or sensory stimuli and measure behavioral responses. And from that, they try to infer how the system works.
And in this case, they infer the properties of neurons just from behavior. And there's like a million variations of this. OK. So now we know that there's neurons in your visual cortex.
The tilt after effect doesn't tell you where in the brain those neurons are. It just says somewhere on your processing chain you have neurons that do that and that adapt out. You need physiology to tell you where.
So now we know that there are neurons in your primary visual cortex that have orientation selectivity. How do you compute that? I keep making all this loose talk about how vision is visual information processing in the youth. You're computing things on representations.
This is actually one of the few cases where there's a pretty good idea of how that's actually computed in a simple neural circuit. So remember that we're going to try to derive this property from a simple circuit starting with the properties of retinal ganglion cells. It's true there's the LGN in between, but the LGN responds much like the retinal ganglion cells. So this is what Hubel and Wiesel proposed.
for which there's some evidence and still some dispute about exactly how this works. But imagine just taking a bunch of those retinal ganglion cells, or I'm sorry, lateral geniculate cells that behave like retinal ganglion cells. Here are four of them. Each of them is a on-center, off-surround spot detector. And if you have them aligned in a row in space, that is, the receptive fields are aligned, not the cells, they respond to different parts of space like this.
And now you have all of them feed into a V1 cell. If it functions as a kind of AND gate, which neurons can do, more or less, then this neuron. is going to detect bars of that orientation. Everybody see how that works? Nice and simple and low tech.
So here's this basic building block in your visual system that you can detect indirectly with adaptation behaviorally, that you can measure neurally. And here we have an idea of how that simple thing is computed. We won't be able to do this for, say, face recognition.
We don't have the circuit for that. But for these simple early building blocks, there are very sensible circuits that can do these first use. computations. All right, so how's this thing going to behave?
Let's imagine a row of these, just as sort of the same thing. But what happens here is if you add up the on-center and the off-surround across those neurons aligned like this, you will get a receptive field of the primary visual cortex neuron that looks like this. Everybody see if you can average that, you get this? OK.
So it has orientation sensitivity, as we just described, but it's also got these flanking fields here, these inhibitory flanking fields here. So if you put stimulus A right in the center like that, it'll turn on like that. You're turning it on in the middle right there.
You get an activation. If you put in a bar right here, right on top of the inhibitory flanker, You're going to get an inhibition in that neuron. And if you put it diagonally, like C, there's no change because the excitation from the center of the field is canceled by the inhibition from the flankers. Everybody get that? So this is just how these are called simple cells, basic orientation selective cells in primary visual cortex.
That's how they behave and how they're computed from the properties of LGN input. Make sense? There's much more to V1. There's all other kinds of selectivities. And we'll skip all that.
Here's the basic idea. OK, so that's one neuron. How are these guys oriented spatially across the brain?
I'm going to go rather quick through a few slides here and then get to some, I hope, more just basic facts. Turns out that they're clustered together. They progress systematically across the cortex.
So here's a piece of cortex, outside the head, inside the head, piece of slab of cortex. What you see is if you send an electrode along the length of cortex, you see this even smooth progression in orientation selectivity. So it's not like random cells. Right next to a cell that likes this, there's a cell that likes that. No, they progress smoothly and evenly across the cortex.
So there's like a little map, a little fine scale map of orientation selectivity. spatially across primary visual cortex. These are sometimes called orientation columns.
And it's another kind of functional organization on top of retinotopy, all in the same chunk of cortex. So primary visual cortex is getting complicated. It isn't just a map. It's a map. And then on top of that map is a smooth progression of orientation happening in the place.
All right. Can we see this with humans? OK, I can do this really fast.
So here's another study with 7 Tesla, super fancy high resolution. Here's a little piece through the back of the brain. Here's the sulcus between the two hemispheres. Here's a piece of V1 in a human subject scanned at 7 Tesla. And in fact, it's claimed that you can see orientation columns like that across the cortex in humans if you have high enough resolution.
Needs to be down to around a millimeter or less. each of those colors is a preferential response to a different orientation. Okay, this can be shown much better in animals, but you can see it here even in humans.