 Technical challenge, shall we say. Okay, so I'm going to talk to you today about understanding attention and predictive coding circuits in the brain. I'm going to put almost all of the emphasis on attention, although I will at least point a direction towards understanding something to do with prediction today. So very importantly, I want to introduce the subject of the research. This research, anything to do with attention requires you to have a behaving animal and we need a behaving animal that we can train and macaque monkeys have proved to be an extremely useful resource for this topic and we use macaques for this research. So everything I'm going to say is based on macaque monkeys. So first of all, I'm going to point out some work done by others. This is Stefan Troyer's research and I just want to go through the concept of attention with you. So we're all familiar with the concept of receptive fields. That is a region in visual space where a neuron is maximally sensitive to something. In this case, if we look at the dashed line here, that is the boundary of a receptive field. So if you put an upwards moving stimulus in that region of space, the neuron that you're recording from is going to respond. The monkey is required to fixate this fixation cross in the center and must not move his or her eyes while they're doing the experiment. And basically, if you present an upward moving stimulus in that receptive field, you get a nice response. Now, if you do something a little bit more clever, if you ask the monkey to attend to the right-hand side stimulus and indicate that he's doing so by pressing a lever on his right-hand side, the response of that neuron will become larger. You'll get a stronger response. If on the other hand, you were to get the monkey to turn his attention to the left-hand side and indicate his attention by indicating with a left-hand lever, you get a much lower response in this neuron. So the stimulus is identical. The only thing that's different is that the monkey is thinking, concentrating or attending, depending which word you want to use, either on the right-hand side or the left-hand side. And this is what we refer to generally as spatial attention. Now, you can do even more complicated and clever experiments, as Stefan's done here. You can actually bias the response of this neuron by always getting the monkey to attend to the left, but you can change the direction of the stimulus on the left-hand side. So in this case, if you ask the monkey to attend to the left, and the left is an upward moving stimulus, which is the correct stimulus for the cell you're recording from, you'll get a strong response, as indicated here in gray. If, on the other hand, you ask the monkey to attend to the left, and it's a downward moving stimulus, and you present an upward moving stimulus in the receptive field, then the response is much lower. So basically, that in a nutshell, as far as neuroscientists are concerned, as opposed to psychologists, defines what we understand by attention. It is enhancing the way the system responds to a sensory input. In my case, the sensory input is always visual, because I'm a vision scientist, but this is equally as effective in other sensory domains. So that's attention. Now, strictly speaking, I'm not working on attention, but what I want to show you is that, using eye movements with behaving monkeys, we can use, we actually can identify some very interesting things that look a lot like attention, and we think this is a good way of being able to investigate attentional mechanisms in monkeys. So let me explain a little bit what I mean. So in order to see, something that we're not really very conscious of most of the time is that we need to look, okay? We don't see by just gazing out into the image in front of us. What we do is we see by constantly moving our eyes around, and this is very evident when you look at this girl's face. You can see, if we measure the eye movements, you can see that the eye movements are spread around the face, pointing that spot of high resolution that we have at the centre of our visual field, the fovea, this area of high resolution here, indicated here in cross-section by these fantastic high-density photoreceptors, which give us that high-resolution vision. We only have good vision in a very small part of our visual field, and we need to move our eyes around in order to get the perception that we have good vision all around. So, where does this all tie in with attention, prediction, and decision-making, which are the three themes of our centre of excellence? So let's have a look at the problem that we face. And by the way, we make one of these saccadic eye movements about three times every second when we're awake. So you're all doing it three times a second right now. You're hardly aware of it. Most of it is purely automated and unconscious. So, here's a picture. If you ask someone to look at that picture for, say, 40 seconds, and you measure their eye movements, this is the kind of eye movement pattern you get. You can see that what people are doing is finding objects of interest, placing the centre of their visual field on that point of interest, and then using their peripheral vision, they make a decision that there's something interesting to look at. They have to attend to the new object. They have to predict where it is, relative to where the eye is pointing now, and they eventually have to make a decision that they're going to move to it. So this is a really nice paradigm to use with the behaving primates, in order to study those three things which form the core of our centre. And the idea here is, of course, that what you're trying to do is... Oops, let's go back. What you're trying to do is constantly present this high-resolution bit of your visual field so that you can recognise the face, but you are simultaneously attending to targets away from that location so that you can make saccades to them. So if we break it down in time, the sequence looks a bit like this. You have approximately 300 milliseconds of stationary image. You fixate that stationary image, in this case. Towards the end of that period, you need to attend to the next target, which is in your peripheral vision. You need to make predictions about where it is and you need to decide to move. Once you've put that in process, you have a saccadic eye movement. Now, something happens during that saccade. Of course, you've got this rapidly moving eye. The image is really poor quality. It's a horrible thing and so we tend to suppress that. And this is commonly referred to as saccadic suppression. I'm not going to talk about that. That's another couple of hours of lectures I could give, but I won't. What I am going to talk about is what happens next. So when you land at the new target, there is a great need, a functional need, to attend very strongly to this new target because anything could have happened. While you're making that saccade, a radical change may have happened in the visual field. Imagine a rugby player trying to catch a ball that's swirling around in the air as it floats about in the wind. They have to make incredibly complex calculations and they have to keep their eye on the ball using saccadic eye movements. And of course, where they predict their saccade is going to go, may be the wrong place because the wind takes the ball in another direction. So they also need to be able to calculate an error. Okay, so now I'm going to start talking about some of the discoveries we've made in the last few years about the way the primate brain deals with this problem and how it does some of the computations that are necessary. So our paradigm in this particular incident was quite simple. So we had one fixation spot, one red fixation spot like this one here, present on a blank screen. The monkeys required to fixate that and they rewarded with fruit juice. We then remove that fixation spot and put another fixation spot some degrees away. And they are required to attend to that target, predict where it is and make a movement to it. And what they do is they create a saccade. This is eye position here and you can see they've made a saccade here. So beautiful fixation, a saccade and then fixation at the new location. What we then did, remember that's on a blank screen. So all that's present is a white screen with one red dot at a time. At random times, relative to those saccades, we presented the monkeys with a fulfilled checkerboard pattern or a random pattern, high contrast pattern. And this was presented for 10 milliseconds. And by the way, we were recording from, in our case, the parietal cortex, the medial superior temporal area of the monkey cortex shown here on your right hand side. So we're recording from this area here. These are the bottom of the electrode tracks. So the monkey's sitting there making saccades, getting rewards and these monkeys would do this for perhaps four hours quite happily before they start to get bored and want to move on, at which point you put them back to bed. So when we present these flash stimuli, the typical response you get from one of these neurons is shown here. So the flash is presented at this point here. And some time later, there's a latent period in the response, there is a response. And the response in this case is a single action potential. And the interesting thing is as you approach the saccade, you'll notice that just prior to and during the saccade, we get a period where that action potential disappears. That's saccadic suppression. What we discovered, and it's a very interesting phenomenon and we didn't believe it when we first saw it, but there's no doubt about it now, many years later and many repetitions later, whenever we presented a visual stimulus from when the saccade lands onwards, we find that the responses are far larger than they occur in the control condition. So before the saccade, the responses are small. Immediately after the saccade, we get this huge boost in the responses to the same stimulus. So to give you a little bit more of a heads up of what that looks like, here is an example. So what I'm doing here is I'm plotting the time prior to the saccade onset. So this is 110 milliseconds before the saccade. The little black object here is the flash stimulus. The cell has some spontaneous activity and then there's a response. Minus 90 milliseconds. Signs of the response are somewhat less and what you see is that by about 50 milliseconds before, we're starting to see just spontaneous activity. Then the saccade happens. We're still getting suppression, saccadic suppression. This response has gone away. The saccade ends and suddenly, if we present stimuli, we can see that these responses are now much larger than they were in the control condition and that continues for some time. So it typically continues through about 150 milliseconds after the saccade and remember, we typically make saccades about every 300 milliseconds. So there's a period of saccadic suppression, which I'm not gonna talk about today and there's this fantastic period of post-saccadic enhancement and we believe that post-saccadic enhancement is related to the attentional phenomena that I showed you earlier. And we believe that it's the same circuitry which drives that attention, also drives this phenomena. So if we look at it in quantitative terms, this is the mean of 75 neurons recorded from macaque cortex. One is the, this is the normalized control response and here is that normalized control response. You can see there's a flash, a latency and then a nice strong response. Coming up to the saccade and during the saccade, there is still a response. You are not blind during saccade, you can see during saccades, but the response is greatly attenuated. Immediately after the saccade, you get an enormous increase in the response and if you remember back to the slide I showed you of Stefan Troyer's work, this looks remarkably like the sort of enhancement you see during attentional shifts and if you think about it, this is an attentional shift. We've asked the monkey to shift his attention from one location in the visual field to another location in the visual field and we believe that what we're seeing here is a long, slow, quite sluggish shift of attention which functionally is offering the animal and indeed does, we're primates too, so we do this. It's offering the visual system an enhanced capacity or an improved sensitivity in the target zone of the saccade. Now we did our experiments in the medial superior temporal area, the parietal cortex of monkeys, that's quite high up in the brain. What's nice is that over the last few years we've started to see a proliferation of people doing these experiments, very similar to the ones I've just described. Here we can see some data from, this is in the same format, so this is the normalized response to the control and this is when the saccade occurs here at zero and you can see saccadic suppression followed by this post-saccadic enhancement and the red line indicates the lateral geniculate nucleus which is a subcortical area which feeds into the cortex. We also see it in the primary visual cortex or V1 which is the green line. We see it in MST which is our data. Another group has repeated those experiments in MST and they get the same result and nicely there's another parietal area, the ventral inter-parietal area which is also showing the same effect. So this seems to be quite a consistent property that you observe in visual cortical areas of the primate brain. What's interesting is someone came along and also had a look in the other major visual pathway which goes through the superior calliculus which is a subcortical area in the brain. It's, I think a lot of people would regard it as the phylogenetically perhaps older area and what you see in the superior calliculus is you get the very strong suppression during the saccade but you do not get this post-saccadic enhancement and that's a really significant finding in terms of circuits because we believe that that visual pathway is treating information really quite differently to the way the cortex is treating its information and the way it's keyed in to the attention mechanisms and keyed into the saccadic control mechanisms. So to have a little look at the circuit then, so here's a macaque brain broken down for you. This is the lateral geniculate nucleus that I talked of and I think I've got the pathways drawn in there. So from the retina into the LGN, from there to the primary visual cortex and up into the parietal cortex here and you can see you get this beautiful post-saccadic effect in all of those areas. The superior calliculus which is another retinal recipient area. It has many connections, I've just drawn two in here. It has a very strong connection with the medial superior temporal area which is one of the areas that we're recording from. It's also heavily connected with this very interesting area at the front of the brain called the frontal eye fields. And the frontal eye fields are interesting because they have cells which fire prior to saccades. So in effect, that's the source of driving our saccadic signals. They also send many strong signals down into the areas that we've been recording from and these signals that come down, they have been strongly associated with attentional mechanisms and they are very important for basically telling the rest of the brain what's about to happen. So we're about to make a saccade and here's the signal telling you what that saccade's gonna look like and where it's gonna go. And we believe it's these signals which have been taken over, if you like, by the sophisticated attentional mechanisms of the brain. In a way, you can think of attention as being a virtual saccade. You are effectively moving your visual field, allocating a different part of the visual field to your maximum interest. So in a nutshell then, we have a circuit so far. We have some interesting pathways. We know that we get this post-cocadic enhancement in the cortical areas and the subcortical areas leading into the cortex, but not in these perhaps evolutionarily older areas. However, there's a problem with all this and I just wanna sort of spend a few minutes just making sure that we're absolutely confident in ascribing the phenomena I've just described to actually ascribing that to an attentional mechanism. So one of the problems is that it is plausible, fully plausible that during this saccade, so sorry, this is eye position. So here's your saccade. This is the same thing differentiated. So this is speed and you can see when the saccade finishes here and that's when you get the post-cocadic enhancement. The problem we have is that it could be that the movement of this saccade across the visual field could do something similar to the sort of thing that Nick told you about this morning. There could be some rapid gain change. There could be a rapid influence of adaptation mechanisms which somehow influences the responses in the post-cocadic period. So it could entirely be explained by those visual consequences. So we had to deal with that and the experiments weren't so easy in order to us to really probe this properly. And the way we chose to do it was very similar to what you've seen already except we didn't worry about anything prior to the saccade which made the experiments a lot easier because we could trigger everything from the saccade. And what we do here is we get the monkey to make a saccade. I don't, for some reason, that's intriguing. Yeah, those are meant to be red lines telling you what's going on there and they've all turned gray so it's almost impossible for you to see them. Sorry about that. That must be to do with changing computers or something. Okay, so what we, I've done it again. What's going on? Yeah, I don't know what's going on. Anyway, so what we do, again, we get the monkey to make a saccade between two fixation spots which you can't see now and then at various times after the saccade ends we remove the fixation spot and we move the image. So the image suddenly moves in the visual field as is meant to be indicated by a large arrow just here. And when it moves in the visual field there is, again, it moves at time. The stimulus moves here at minus 50. Sorry, no it doesn't. It moves at zero and there's a latent period and again we're recording from neurons in MST and you can see there's a strong response with a latency of around about 45 milliseconds. And because we've removed the fixation spot when we move that image, the monkey reflexively tracks it. So there is a reflexive tracking movement and you can see the eye movements here where the monkey's actually chasing the stimulus. And that's really useful because we get a behavioral measure from that. So we're recording the neural responses and we're simultaneously recording the animal's reflexive behavior. Now unfortunately we're not also measuring the animal's perception. That would be wonderful if we could do that simultaneously. But at least we've got a couple of measures here which we can work off. And what we find, which ties in beautifully with what you've already seen, is that if the motion stimulus occurs very soon after the cicade has ended, we get these really robust strong responses. If however, we make the monkey fixate for a long time so that the influence of the cicade goes away, you can see that the response is far smaller, which is the gray line. So again, this is post-cicadic enhancement that you're seeing here. So that's kind of a repeat of what you've already seen. Interestingly, down here you can see the eye speed traces and you see the same phenomena with the eye speed trace. When we get the monkey to track a moving scene very soon after a cicade, you get very high initial eye speeds. When we ask them to do it a long time after a cicade, the eye speed is slower. So this mechanism we're seeing is enhancing the ability to track images and it's generating stronger responses in the brain immediately after cicades. But what we needed to do is make sure that it wasn't just the movement that occurred during the cicade that was causing these effects. So what we did was we got the monkeys to rather than make a cicade, we got them to fixate and we made the visual movement for them. So we replicated what they saw during a cicade and we made simulated cicades. And what we found, which was really very interesting indeed, quite unexpected too, was that what you've got here is a simulated cicade in green, sorry, a motion stimulus which is occurring immediately after a simulated cicade. And you can see in the gray, a motion stimulus that's occurring a long time after a cicade and they're identical. The two responses are the same whereas they are not the same in this situation here. So it's quite clear that it is absolutely essential for the monkey to actually make the decision to make the cicade, to see the kind of enhancement of visual responses that we see in the post-cicadic period. You can't trick the monkey, you can't trick the system. The system knows whether it's made the choice and sent those signals and enhanced the signals. Now this is the same, there's nothing new in this slide except that this is all of our data together. So in the top, you can see what we call the short delay situation is enhanced in comparison to the long delay. So when you are in the post-cicadic period, you get a stronger response than when you are not. For simulated cicades, this is not the case. Across a large population of neurons and in the bottom, you see the same. You still see this enhancement and what we did in this experiment was actually slightly more clever. When the monkeys were making a cicade, we actually removed the visual stimulus during the cicade. Quite a tricky experiment to do and there's lots of controls we had to do to get these experiments right. But again, it demonstrates very nicely that you get post-cicadic enhancement that is not related to the vision that occurs during the cicade. It is an internally generated problem. So in summary, this is my last slide, we've reached a really nice point where we have a lot of information about the visual pathway. We have found a really interesting phenomenon that immediately after cicades, we get this very significant increase in the visual sensitivity of neurons throughout the cortical pathway of the brain. Interestingly, we don't see that in the superior colliculus pathway, suggesting that there really is quite a different mechanism. And we believe that, and it's what we're working on now to try and confirm this, but we believe that this is because the connections that the frontal eye field makes with these areas, both cortical areas and the feed-through to those cortical areas, we believe that the frontal eye field is sending signals into those areas, alerting those areas to the fact that they have to attend very strongly to a new target in the visual field and offering the cells a dramatic increase in their sensitivity to the visual stimuli that are presented during that period. And our intention now is to work on these circuits, we can do a number of experiments, we can block these pathways. The nice thing is that the pathways that come from the frontal eye fields that actually drive the eye movements are different from the pathways which send signals into these cortical areas. So we can block these cortical areas without affecting the monkey's ability to make cicades. And that's our intention, is to move along in that way and use specific blockades of these pathways to try and work out what's going on from a mechanistic point of view. Thank you. Thank you very much, Michael. We have time for a couple of questions. Well, I've got lots of questions. I might offer David a chance here and then John at the back. You kind of begged a question when you mentioned that you can't ask the monkeys how they perceive things. So an obvious step would be to repeat these experiments in humans. Obviously the ethics committees won't let you stick needles in, but you can get the behavioural... Yes, so we... ...the same and then ask questions. We get the same enhancement effects happening in humans. So we've done... In fact, in the last three years, we've been running... We want to get a lot of subjects because we really want to make the finding robust. So in the last three years, we've had cohorts of students come through and we can actually see that there is a very clear post-cocatic enhancement in human perception that follows the Cicade. It's very specific, which is really interesting. So it depends on the visual environment quite strongly. So clearly, I think we... Perhaps we were a little lucky and we chose a really good visual environment for the monkeys, but with the human observers, we find that the visual environment is really critical. In certain visual environments, you get very powerful enhancement and in other visual environments, you get less enhancement. Interesting. Hi, Mike. That was great, Michael. I just wondered if the... You didn't really show this data, but is the post-cocatic enhancement value-sensitive of the stimulus that you're finally circuiting to? So is it sensitive to distractors? And the reason I asked that is that... You might know Mitsumoto and Takata did some work recently where they were looking at visual search and recording from dopamine cells and they found that when they had specific stimulus oriented in a way amongst distractors, the longer they searched, when they found the thing, they had a much larger post-cocatic response, which was the first time it's been described in the cells. And it's obviously too slow for Calculus. So I wondered, do you see enhancement in the... So we haven't specifically looked at that, but the answer to your question, I think, is almost certainly in the affirmative. Yeah, I think it's going to be very specific in the same way that the attentional mechanisms are. So there's these lovely experiments that have been done where the orientation of the distractor or the other feature really has a major impact on what happens. We haven't had a chance to do those experiments because they're very time consuming, but I think that'd be a really exciting direction to go in. Cheers. Michael, that's very interesting stuff. The post-cocatic suppression and the post-cocatic enhancement after suppression, I wonder what are the best arguments to really relate it to attention and not to interpret it as a post-inhibitory rebound, which is found in many visual receptive fields in regions where you have had strong inhibition. You get also an enhancement like this. And related to this, my question would be, do you have psychotic suppression in the Calculus? Yes, yes, there's strong psychotic suppression in the Calculus. Okay, so that would be an argument against my thinking. And the answer to your question actually is very strongly related to the question that came before. So we haven't done it yet, but the way to really get to the issue that this is a specific attentional effect is to show that it's very specific to the particular attentional needs of that task. So we've used a fairly simple task at this time, and I think that we'll use more sophisticated stimuli to probe that. But I think certainly our initial experiments suggest that if you alter the stimulus so that you're presenting like things, you get very strong effects if they're dislike things, you don't. And that suggests to us that it's got a strong relation to attention. Okay, thank you very much, Michael. We have to move on. So thanks again, Michael, for your presentation.