WEBVTT 00:00:00.000 --> 00:00:01.979 align:middle line:90% [SQUEAKING] 00:00:01.979 --> 00:00:03.458 align:middle line:90% [RUSTLING] 00:00:03.458 --> 00:00:04.937 align:middle line:90% [CLICKING] 00:00:04.937 --> 00:00:09.215 align:middle line:90% 00:00:09.215 --> 00:00:10.840 align:middle line:84% JOSH MCDERMOTT: Let me just kick things 00:00:10.840 --> 00:00:13.400 align:middle line:90% off by showing you a picture. 00:00:13.400 --> 00:00:20.840 align:middle line:84% So this is what is called a flat map of the macaque brain. 00:00:20.840 --> 00:00:26.080 align:middle line:84% So the cortex is this sheet which, in your head, 00:00:26.080 --> 00:00:29.820 align:middle line:84% is kind of folded up into the shape of your brain, basically. 00:00:29.820 --> 00:00:34.540 align:middle line:84% And a flat map consists of taking the cortex, 00:00:34.540 --> 00:00:37.800 align:middle line:84% making a few cuts, and then kind of flattening things out, 00:00:37.800 --> 00:00:40.700 align:middle line:84% such that you can see the entire cortical surface at once. 00:00:40.700 --> 00:00:42.520 align:middle line:90% All right. 00:00:42.520 --> 00:00:45.800 align:middle line:84% And this is a flat map where what 00:00:45.800 --> 00:00:49.080 align:middle line:84% are considered to be different regions of the visual system 00:00:49.080 --> 00:00:53.133 align:middle line:84% are outlined, and different areas are in different colors. 00:00:53.133 --> 00:00:54.800 align:middle line:84% All of the regions here that are colored 00:00:54.800 --> 00:00:58.360 align:middle line:84% are considered to be part of the visual system. 00:00:58.360 --> 00:00:59.980 align:middle line:84% That means that they respond when 00:00:59.980 --> 00:01:01.580 align:middle line:90% people are looking at stuff. 00:01:01.580 --> 00:01:04.143 align:middle line:84% And so this is the intact macaque brain. 00:01:04.143 --> 00:01:06.060 align:middle line:84% And so you can see that that the visual system 00:01:06.060 --> 00:01:09.660 align:middle line:84% is the back half of the brain, along with some stuff here 00:01:09.660 --> 00:01:10.725 align:middle line:90% in the frontal lobe. 00:01:10.725 --> 00:01:12.100 align:middle line:84% And this is kind of what it looks 00:01:12.100 --> 00:01:14.140 align:middle line:90% like when you flatten it out. 00:01:14.140 --> 00:01:16.800 align:middle line:84% And so there's a lot of different regions, 00:01:16.800 --> 00:01:18.220 align:middle line:84% and it's a big chunk of the brain. 00:01:18.220 --> 00:01:18.720 align:middle line:90% All right. 00:01:18.720 --> 00:01:22.420 align:middle line:84% So what is it that distinguishes the different visual areas? 00:01:22.420 --> 00:01:25.820 align:middle line:84% Well, the main criteria by which you typically 00:01:25.820 --> 00:01:29.420 align:middle line:84% distinguish visual areas is retinotopy, 00:01:29.420 --> 00:01:34.380 align:middle line:84% where each visual area typically contains a retinotopic map. 00:01:34.380 --> 00:01:37.860 align:middle line:84% So a map of the entire visual field. 00:01:37.860 --> 00:01:41.660 align:middle line:84% And so the visual areas can be localized by measuring 00:01:41.660 --> 00:01:42.940 align:middle line:90% retinotopy. 00:01:42.940 --> 00:01:45.740 align:middle line:84% In this case, by measuring how the neural responses 00:01:45.740 --> 00:01:47.880 align:middle line:90% change with stimulus location. 00:01:47.880 --> 00:01:52.380 align:middle line:84% So this is an example where a participant in an experiment 00:01:52.380 --> 00:01:54.240 align:middle line:90% is looking at this display. 00:01:54.240 --> 00:01:56.380 align:middle line:84% So they stare at the fixation point. 00:01:56.380 --> 00:01:59.970 align:middle line:84% And there are these annuluses that kind of gradually expand. 00:01:59.970 --> 00:02:01.350 align:middle line:84% So there are checkerboards, which 00:02:01.350 --> 00:02:02.933 align:middle line:84% are these high-contrast stimuli, which 00:02:02.933 --> 00:02:04.290 align:middle line:90% produce big visual responses. 00:02:04.290 --> 00:02:06.650 align:middle line:90% So the idea is that it expands. 00:02:06.650 --> 00:02:09.470 align:middle line:84% And so the location of the visual stimulus 00:02:09.470 --> 00:02:13.430 align:middle line:84% varies over eccentricity as a function of time. 00:02:13.430 --> 00:02:15.110 align:middle line:84% And in this case, you have a pie wedge 00:02:15.110 --> 00:02:17.370 align:middle line:84% of a checkerboard that rotates around. 00:02:17.370 --> 00:02:20.870 align:middle line:90% So you map out polar angle. 00:02:20.870 --> 00:02:25.630 align:middle line:84% And so in each case, every point in the visual cortex 00:02:25.630 --> 00:02:28.310 align:middle line:84% is color-coded, in this case, as a function 00:02:28.310 --> 00:02:32.270 align:middle line:84% of the eccentricity at which the stimulus evokes the biggest 00:02:32.270 --> 00:02:33.010 align:middle line:90% response. 00:02:33.010 --> 00:02:36.410 align:middle line:84% And so you can see that you get this kind of gradient here. 00:02:36.410 --> 00:02:38.452 align:middle line:84% And in this case, as a function of polar angle. 00:02:38.452 --> 00:02:39.910 align:middle line:84% And you get another gradient that's 00:02:39.910 --> 00:02:41.830 align:middle line:90% kind of orthogonal to this. 00:02:41.830 --> 00:02:44.890 align:middle line:84% So this is the retinotopic map that we have seen before. 00:02:44.890 --> 00:02:46.750 align:middle line:84% So we talked a lot about this in the context 00:02:46.750 --> 00:02:48.790 align:middle line:90% of primary visual cortex. 00:02:48.790 --> 00:02:53.670 align:middle line:84% So remember this diagram where this 00:02:53.670 --> 00:02:57.290 align:middle line:84% describes how the two visual fields get mapped 00:02:57.290 --> 00:02:58.790 align:middle line:90% onto your visual system. 00:02:58.790 --> 00:03:01.810 align:middle line:84% So each eye can see part of both visual fields, 00:03:01.810 --> 00:03:05.250 align:middle line:84% but then the contralateral visual field kind of gets-- 00:03:05.250 --> 00:03:07.290 align:middle line:84% or I should say, the left visual field 00:03:07.290 --> 00:03:09.047 align:middle line:84% gets mapped to the right hemisphere, 00:03:09.047 --> 00:03:11.630 align:middle line:84% and the right visual field gets mapped to the left hemisphere. 00:03:11.630 --> 00:03:15.710 align:middle line:90% 00:03:15.710 --> 00:03:19.210 align:middle line:84% So this is an example of an experiment 00:03:19.210 --> 00:03:22.190 align:middle line:84% where you can see a whole bunch of different visual regions. 00:03:22.190 --> 00:03:24.370 align:middle line:90% So this is a one-- 00:03:24.370 --> 00:03:26.170 align:middle line:90% this is from one hemisphere. 00:03:26.170 --> 00:03:29.650 align:middle line:84% And here, the color is mapping out 00:03:29.650 --> 00:03:33.010 align:middle line:84% the polar angle at which the stimulus evoked the biggest 00:03:33.010 --> 00:03:34.050 align:middle line:90% response. 00:03:34.050 --> 00:03:36.650 align:middle line:84% And so you can see in what's considered 00:03:36.650 --> 00:03:40.410 align:middle line:84% to be area V1, primary visual cortex, the color kind of sweeps 00:03:40.410 --> 00:03:44.170 align:middle line:84% from green to red all the way up to dark blue. 00:03:44.170 --> 00:03:45.710 align:middle line:90% And so this is one hemisphere. 00:03:45.710 --> 00:03:48.590 align:middle line:84% And so you're only getting one hemifield, 00:03:48.590 --> 00:03:52.850 align:middle line:84% but you see the full range of visual angles. 00:03:52.850 --> 00:03:55.450 align:middle line:84% So then what happens, once you get 00:03:55.450 --> 00:03:58.630 align:middle line:84% to what's considered to be the border of V2, is you 00:03:58.630 --> 00:04:00.335 align:middle line:84% start to see a lot of blue and green. 00:04:00.335 --> 00:04:02.710 align:middle line:84% So there are these gradients, but they're blue and green, 00:04:02.710 --> 00:04:04.168 align:middle line:84% and there's not a whole lot of red, 00:04:04.168 --> 00:04:09.790 align:middle line:84% whereas below V1 you see gradients from red to green. 00:04:09.790 --> 00:04:13.270 align:middle line:84% And so this is an indication that there's 00:04:13.270 --> 00:04:18.269 align:middle line:84% kind of a separation of the upper and the lower hemifields. 00:04:18.269 --> 00:04:19.490 align:middle line:90% But they're inverted. 00:04:19.490 --> 00:04:24.290 align:middle line:84% So the lower hemifield is mapped onto the region above V1, 00:04:24.290 --> 00:04:27.670 align:middle line:84% and the upper hemifield is mapped onto the region below V1. 00:04:27.670 --> 00:04:29.990 align:middle line:84% And so these visual areas have been 00:04:29.990 --> 00:04:33.790 align:middle line:84% given names that are a little bit complicated 00:04:33.790 --> 00:04:37.990 align:middle line:84% sometimes because of the nonlinear historical trajectory 00:04:37.990 --> 00:04:39.530 align:middle line:90% over which they were discovered. 00:04:39.530 --> 00:04:42.710 align:middle line:84% But you've got V1, V2, V3, and then 00:04:42.710 --> 00:04:45.510 align:middle line:84% there's this other one called V3a, because they had already 00:04:45.510 --> 00:04:47.390 align:middle line:84% discovered V4, and then they realized 00:04:47.390 --> 00:04:49.807 align:middle line:84% there was this extra region and they had to squeeze it in. 00:04:49.807 --> 00:04:52.030 align:middle line:90% So there's stuff like that. 00:04:52.030 --> 00:04:52.530 align:middle line:90% All right. 00:04:52.530 --> 00:04:53.905 align:middle line:84% So there's lots of these regions. 00:04:53.905 --> 00:04:55.850 align:middle line:84% The key idea is that we distinguish 00:04:55.850 --> 00:04:58.970 align:middle line:84% these different regions via retinotopy. 00:04:58.970 --> 00:05:00.890 align:middle line:84% And so, in particular, what happens 00:05:00.890 --> 00:05:03.310 align:middle line:84% when you get to the border of a visual region 00:05:03.310 --> 00:05:05.930 align:middle line:84% is that the sign of the retinotopy flips. 00:05:05.930 --> 00:05:08.150 align:middle line:84% So here it's going from green to red. 00:05:08.150 --> 00:05:11.210 align:middle line:84% And then it switches back to going from red to blue. 00:05:11.210 --> 00:05:13.330 align:middle line:84% And the same thing happens with tonotopy 00:05:13.330 --> 00:05:14.770 align:middle line:84% in the auditory system, where you 00:05:14.770 --> 00:05:18.810 align:middle line:84% see a reversal of the sign of the map. 00:05:18.810 --> 00:05:21.318 align:middle line:84% And so in the early days, long before there was fMRI 00:05:21.318 --> 00:05:23.110 align:middle line:84% and you could look at this stuff in humans, 00:05:23.110 --> 00:05:26.290 align:middle line:84% people were doing this in non-human animals 00:05:26.290 --> 00:05:27.430 align:middle line:90% with electrodes. 00:05:27.430 --> 00:05:29.575 align:middle line:84% So they'd have electrodes in the brain. 00:05:29.575 --> 00:05:31.450 align:middle line:84% They'd be measuring receptive field locations 00:05:31.450 --> 00:05:33.770 align:middle line:84% and moving the electrode around and observing 00:05:33.770 --> 00:05:35.490 align:middle line:84% how the location of the receptive 00:05:35.490 --> 00:05:37.630 align:middle line:84% fields kind of varied across the cortical surface. 00:05:37.630 --> 00:05:39.297 align:middle line:84% But the same kind of principles applied. 00:05:39.297 --> 00:05:42.670 align:middle line:84% It's just with fMRI, you can look at everything at once. 00:05:42.670 --> 00:05:47.010 align:middle line:84% And so it's helpful for looking at these maps in this way. 00:05:47.010 --> 00:05:50.730 align:middle line:84% So another kind of important concept that we've already 00:05:50.730 --> 00:05:54.470 align:middle line:84% alluded to is this idea that you can 00:05:54.470 --> 00:05:57.510 align:middle line:84% think about the visual system as consisting of pathways, 00:05:57.510 --> 00:06:00.390 align:middle line:84% and that these pathways kind of have their roots in the retina. 00:06:00.390 --> 00:06:03.270 align:middle line:84% So we've talked about how you've got 00:06:03.270 --> 00:06:06.710 align:middle line:84% midget cells and parasol cells and that they 00:06:06.710 --> 00:06:09.430 align:middle line:84% project to these different parts of the LGN. 00:06:09.430 --> 00:06:12.470 align:middle line:84% Now, OK, this diagram has this other slightly annoying 00:06:12.470 --> 00:06:15.670 align:middle line:84% complexity to this, which is that occasionally people 00:06:15.670 --> 00:06:18.830 align:middle line:84% change the nomenclature with which retinal ganglion cells are 00:06:18.830 --> 00:06:19.650 align:middle line:90% described. 00:06:19.650 --> 00:06:22.910 align:middle line:84% And so sometimes the parasol cells 00:06:22.910 --> 00:06:25.910 align:middle line:90% are referred to as M cells. 00:06:25.910 --> 00:06:27.870 align:middle line:84% I guess they thought this would make it easier 00:06:27.870 --> 00:06:31.150 align:middle line:84% to remember that they project to the magnocellular layers. 00:06:31.150 --> 00:06:34.250 align:middle line:84% And sometimes the midget cells are referred to as P cells. 00:06:34.250 --> 00:06:36.770 align:middle line:90% So it's just a gigantic mess. 00:06:36.770 --> 00:06:38.650 align:middle line:90% But you'll all keep it straight. 00:06:38.650 --> 00:06:40.070 align:middle line:84% But the essential idea is that you 00:06:40.070 --> 00:06:41.570 align:middle line:84% have these two different populations 00:06:41.570 --> 00:06:43.250 align:middle line:90% of neurons in the retina. 00:06:43.250 --> 00:06:46.010 align:middle line:90% They have different properties. 00:06:46.010 --> 00:06:48.470 align:middle line:84% And they project to different layers of the LGN. 00:06:48.470 --> 00:06:50.430 align:middle line:84% Those in turn project to different parts 00:06:50.430 --> 00:06:53.560 align:middle line:84% of primary visual cortex, these different sublayers. 00:06:53.560 --> 00:06:56.560 align:middle line:84% And then those in turn project to different parts 00:06:56.560 --> 00:06:58.020 align:middle line:90% of subsequent areas. 00:06:58.020 --> 00:07:01.920 align:middle line:84% So area 18, which is often known as V2, 00:07:01.920 --> 00:07:04.033 align:middle line:84% has these different subcomponents to it. 00:07:04.033 --> 00:07:06.200 align:middle line:84% And so there's this degree of functional segregation 00:07:06.200 --> 00:07:07.920 align:middle line:90% that kind of persists. 00:07:07.920 --> 00:07:12.160 align:middle line:84% And it persists even up to areas that are 00:07:12.160 --> 00:07:14.260 align:middle line:90% fairly deep in the hierarchy. 00:07:14.260 --> 00:07:19.760 align:middle line:84% So inferior temporal cortex and the parietal lobe. 00:07:19.760 --> 00:07:26.160 align:middle line:84% And so one way to think about this organization that remains 00:07:26.160 --> 00:07:29.240 align:middle line:84% very common and probably, to some extent, true, 00:07:29.240 --> 00:07:33.040 align:middle line:84% is this idea that you can think of two main pathways 00:07:33.040 --> 00:07:38.400 align:middle line:84% in the visual system-- one that extends ventrally, 00:07:38.400 --> 00:07:40.500 align:middle line:84% that mediates object recognition, 00:07:40.500 --> 00:07:43.100 align:middle line:84% culminating in inferotemporal cortex-- 00:07:43.100 --> 00:07:44.760 align:middle line:84% it's often called the "what" pathway-- 00:07:44.760 --> 00:07:47.680 align:middle line:84% and then one extending dorsally, culminating in the parietal 00:07:47.680 --> 00:07:50.620 align:middle line:84% lobe, that's involved in the localization of objects, 00:07:50.620 --> 00:07:54.700 align:middle line:84% often for the purpose of mediating actions. 00:07:54.700 --> 00:07:57.980 align:middle line:84% But to some extent, this division 00:07:57.980 --> 00:08:01.820 align:middle line:84% has its roots earlier in the visual system. 00:08:01.820 --> 00:08:04.220 align:middle line:84% So this is a really famous picture. 00:08:04.220 --> 00:08:06.840 align:middle line:84% It kind of looks like a horrible subway map, 00:08:06.840 --> 00:08:10.480 align:middle line:84% but it's actually a diagram of the primate visual system. 00:08:10.480 --> 00:08:13.840 align:middle line:84% So each little box on this picture is a visual area. 00:08:13.840 --> 00:08:15.780 align:middle line:90% So distinguished by retinotopy. 00:08:15.780 --> 00:08:18.900 align:middle line:84% And then the lines between them represent connections, 00:08:18.900 --> 00:08:21.360 align:middle line:84% or I guess, fairly dense connections, 00:08:21.360 --> 00:08:23.140 align:middle line:84% dense enough that they considered it worth 00:08:23.140 --> 00:08:24.380 align:middle line:90% plotting on the graph. 00:08:24.380 --> 00:08:27.580 align:middle line:84% So these are the retinal ganglion cells. 00:08:27.580 --> 00:08:30.060 align:middle line:84% You've got these two classes predominantly, 00:08:30.060 --> 00:08:33.260 align:middle line:84% the two different layers, types of layers of the LGN, 00:08:33.260 --> 00:08:37.500 align:middle line:84% V1, V2, and then an assortment of other areas. 00:08:37.500 --> 00:08:40.740 align:middle line:84% And so you can see that the things to take away 00:08:40.740 --> 00:08:44.059 align:middle line:84% from this are, one, there's a lot of areas. 00:08:44.059 --> 00:08:46.300 align:middle line:84% Two, there's a lot of connections. 00:08:46.300 --> 00:08:52.260 align:middle line:84% Three, these are organized and conceptualized in a hierarchy. 00:08:52.260 --> 00:08:54.800 align:middle line:84% So we start out at the beginning of the system where 00:08:54.800 --> 00:08:56.120 align:middle line:90% the light enters. 00:08:56.120 --> 00:09:01.360 align:middle line:84% And then some of these regions are kind of situated more deeply 00:09:01.360 --> 00:09:03.300 align:middle line:90% into the system than others. 00:09:03.300 --> 00:09:08.120 align:middle line:84% So that's a very important idea, is that the sensory system 00:09:08.120 --> 00:09:09.300 align:middle line:90% is hierarchical. 00:09:09.300 --> 00:09:11.880 align:middle line:84% So some regions are getting input from others 00:09:11.880 --> 00:09:15.760 align:middle line:84% and then providing input to others. 00:09:15.760 --> 00:09:20.733 align:middle line:84% And look at what happens when you move 00:09:20.733 --> 00:09:21.900 align:middle line:90% more deeply into the system. 00:09:21.900 --> 00:09:24.400 align:middle line:84% So if you record from neurons in these different areas 00:09:24.400 --> 00:09:26.640 align:middle line:84% and try to understand what they represent 00:09:26.640 --> 00:09:29.880 align:middle line:84% or what they're responding to, again, another common theme 00:09:29.880 --> 00:09:32.580 align:middle line:84% is that as you move more deep into the system, 00:09:32.580 --> 00:09:34.500 align:middle line:84% the responses get more complicated. 00:09:34.500 --> 00:09:35.357 align:middle line:90% All right. 00:09:35.357 --> 00:09:36.940 align:middle line:84% And so there are very famous examples, 00:09:36.940 --> 00:09:38.898 align:middle line:84% so famous that you've probably encountered them 00:09:38.898 --> 00:09:39.660 align:middle line:90% in other classes. 00:09:39.660 --> 00:09:42.960 align:middle line:84% So this is an example of a neuron in inferotemporal cortex 00:09:42.960 --> 00:09:45.240 align:middle line:90% that's selective for faces. 00:09:45.240 --> 00:09:47.440 align:middle line:84% So it will respond a lot when you present it 00:09:47.440 --> 00:09:50.460 align:middle line:90% with an image of a face. 00:09:50.460 --> 00:09:52.260 align:middle line:84% Less so when you blur out the eyes. 00:09:52.260 --> 00:09:54.340 align:middle line:84% It can even be kind of a schematic of a face, 00:09:54.340 --> 00:09:56.940 align:middle line:84% and you get a pretty good size response. 00:09:56.940 --> 00:10:00.163 align:middle line:84% So these neurons are much harder, much 00:10:00.163 --> 00:10:02.580 align:middle line:84% more difficult to describe mathematically than the neurons 00:10:02.580 --> 00:10:04.820 align:middle line:90% that you see in V1. 00:10:04.820 --> 00:10:07.740 align:middle line:84% But often selective to more complicated things 00:10:07.740 --> 00:10:09.740 align:middle line:84% that are more behaviorally relevant. 00:10:09.740 --> 00:10:11.700 align:middle line:84% And so there are some pretty famous examples 00:10:11.700 --> 00:10:14.800 align:middle line:84% at this point that really illustrate this point. 00:10:14.800 --> 00:10:18.520 align:middle line:84% So how many people have heard of the Jennifer Aniston neuron? 00:10:18.520 --> 00:10:19.020 align:middle line:90% Yeah. 00:10:19.020 --> 00:10:22.340 align:middle line:84% So that's one that made its way into pop culture. 00:10:22.340 --> 00:10:26.380 align:middle line:84% So this is in inferotemporal cortex of a human who evidently 00:10:26.380 --> 00:10:29.260 align:middle line:90% watched some TV or movies. 00:10:29.260 --> 00:10:31.440 align:middle line:90% And you can see-- 00:10:31.440 --> 00:10:35.260 align:middle line:84% so the diagram here shows example images and then, 00:10:35.260 --> 00:10:38.300 align:middle line:84% underneath it, the response of this particular cell. 00:10:38.300 --> 00:10:41.540 align:middle line:84% And so you can see that there are these two different images 00:10:41.540 --> 00:10:42.520 align:middle line:90% of Jennifer Aniston. 00:10:42.520 --> 00:10:45.540 align:middle line:84% They both produce a pretty good sized response. 00:10:45.540 --> 00:10:48.330 align:middle line:84% This is kind of non-trivial because the images, 00:10:48.330 --> 00:10:49.970 align:middle line:84% if you look at the pixel intensities, 00:10:49.970 --> 00:10:51.910 align:middle line:84% they don't really have a whole lot in common. 00:10:51.910 --> 00:10:53.550 align:middle line:84% Even in abstract terms, I mean, there's 00:10:53.550 --> 00:10:57.310 align:middle line:84% glasses being worn in one case and not the other. 00:10:57.310 --> 00:10:59.990 align:middle line:84% There's just lots of differences. 00:10:59.990 --> 00:11:02.010 align:middle line:84% And it doesn't respond to lots of other images. 00:11:02.010 --> 00:11:07.110 align:middle line:84% Looks like Brad Pitt decreases the response a little bit. 00:11:07.110 --> 00:11:09.350 align:middle line:84% You see stuff like this deep in the system. 00:11:09.350 --> 00:11:11.310 align:middle line:90% This is my favorite example. 00:11:11.310 --> 00:11:14.508 align:middle line:84% This is a neuron in entorhinal cortex. 00:11:14.508 --> 00:11:16.050 align:middle line:84% I don't know how easy this is to see, 00:11:16.050 --> 00:11:18.310 align:middle line:84% but this is somebody who evidently liked Star Wars 00:11:18.310 --> 00:11:22.130 align:middle line:84% because the neuron responds to images of Luke Skywalker, 00:11:22.130 --> 00:11:26.890 align:middle line:84% but it also responds to the words "Luke Skywalker" in text. 00:11:26.890 --> 00:11:29.390 align:middle line:84% It also responds to someone saying the words "Luke 00:11:29.390 --> 00:11:30.870 align:middle line:90% Skywalker." 00:11:30.870 --> 00:11:35.550 align:middle line:84% And there's a moderate response to Yoda. 00:11:35.550 --> 00:11:39.615 align:middle line:84% So you move deep into these sensory systems and things 00:11:39.615 --> 00:11:40.490 align:middle line:90% get more complicated. 00:11:40.490 --> 00:11:41.532 align:middle line:90% So that's a common theme. 00:11:41.532 --> 00:11:45.370 align:middle line:84% And we'll elaborate on that in much more detail and much more 00:11:45.370 --> 00:11:47.750 align:middle line:90% rigor in subsequent lectures. 00:11:47.750 --> 00:11:50.610 align:middle line:90% 00:11:50.610 --> 00:11:54.090 align:middle line:84% And a lot of times, one of the other things 00:11:54.090 --> 00:11:58.210 align:middle line:84% that you often see signs of as you move more 00:11:58.210 --> 00:12:00.810 align:middle line:84% deeply into sensory systems is that the responses 00:12:00.810 --> 00:12:02.313 align:middle line:90% become more invariant. 00:12:02.313 --> 00:12:04.230 align:middle line:84% And remember, at the very start of the course, 00:12:04.230 --> 00:12:07.570 align:middle line:84% we talked about one of the challenges of, in particular, 00:12:07.570 --> 00:12:09.410 align:middle line:84% recognition tasks-- so one of the things 00:12:09.410 --> 00:12:12.390 align:middle line:84% our sensory systems are good for is helping us recognize things. 00:12:12.390 --> 00:12:14.770 align:middle line:84% And one of the challenges of recognition tasks 00:12:14.770 --> 00:12:18.770 align:middle line:84% is that different images or different sounds 00:12:18.770 --> 00:12:21.970 align:middle line:84% of the same thing in the world can often physically 00:12:21.970 --> 00:12:23.530 align:middle line:90% be really different. 00:12:23.530 --> 00:12:27.810 align:middle line:84% And so these are all different images of a house. 00:12:27.810 --> 00:12:30.170 align:middle line:84% There's in fact an image of the same house taken 00:12:30.170 --> 00:12:31.770 align:middle line:90% from two different viewpoints. 00:12:31.770 --> 00:12:34.270 align:middle line:84% So you can both tell that these things are all houses. 00:12:34.270 --> 00:12:36.670 align:middle line:84% You can also pick out which two are the exact same house. 00:12:36.670 --> 00:12:38.170 align:middle line:84% But of course, the pixel intensities 00:12:38.170 --> 00:12:40.970 align:middle line:84% are completely different in all of these different cases. 00:12:40.970 --> 00:12:45.030 align:middle line:84% So somehow or another, you have to construct representations 00:12:45.030 --> 00:12:48.390 align:middle line:90% where those invariances exist. 00:12:48.390 --> 00:12:50.630 align:middle line:84% Now, what we're going to talk about today is not 00:12:50.630 --> 00:12:52.390 align:middle line:84% this kind of stuff so much, but something 00:12:52.390 --> 00:12:56.830 align:middle line:84% that's kind of in between these very complicated recognition 00:12:56.830 --> 00:12:59.610 align:middle line:84% tasks and what we call early vision. 00:12:59.610 --> 00:13:04.030 align:middle line:84% And this normally is what is referred to as mid-level vision. 00:13:04.030 --> 00:13:06.030 align:middle line:84% So remember, the big picture here 00:13:06.030 --> 00:13:08.270 align:middle line:84% is that perception involves inferring 00:13:08.270 --> 00:13:11.510 align:middle line:84% the structure of the world from measurements of energy that 00:13:11.510 --> 00:13:13.550 align:middle line:90% are generated by the world. 00:13:13.550 --> 00:13:17.270 align:middle line:84% So in the case of vision, these are patterns of light. 00:13:17.270 --> 00:13:20.567 align:middle line:84% And so far, the past three or four lectures, 00:13:20.567 --> 00:13:22.150 align:middle line:84% we've been talking about what is often 00:13:22.150 --> 00:13:23.690 align:middle line:90% referred to as early vision. 00:13:23.690 --> 00:13:25.710 align:middle line:84% So early vision we often think of as like a set 00:13:25.710 --> 00:13:27.390 align:middle line:90% of useful measurements. 00:13:27.390 --> 00:13:30.910 align:middle line:84% So you get this image as an input to the retina. 00:13:30.910 --> 00:13:33.172 align:middle line:84% And then there are a set of filters 00:13:33.172 --> 00:13:34.630 align:middle line:84% that are measuring different things 00:13:34.630 --> 00:13:37.050 align:middle line:84% about the image in different positions in space. 00:13:37.050 --> 00:13:39.113 align:middle line:84% So in V1, for instance, the various kinds 00:13:39.113 --> 00:13:41.030 align:middle line:84% of receptive fields give us local measurements 00:13:41.030 --> 00:13:44.380 align:middle line:84% of orientation, contrast, disparity, color, 00:13:44.380 --> 00:13:47.400 align:middle line:84% spatial frequency, and so forth, sort of like the ingredients 00:13:47.400 --> 00:13:49.880 align:middle line:90% of images. 00:13:49.880 --> 00:13:51.650 align:middle line:84% Now there are also perceptual phenomena 00:13:51.650 --> 00:13:53.400 align:middle line:84% that are linked to these measurements that 00:13:53.400 --> 00:13:56.420 align:middle line:84% also fall under what's called early vision. 00:13:56.420 --> 00:13:59.320 align:middle line:84% And those tend to be phenomena that people classically 00:13:59.320 --> 00:14:02.180 align:middle line:84% think are explained by those measurements. 00:14:02.180 --> 00:14:05.520 align:middle line:84% So for instance, we talked about these adaptation effects, 00:14:05.520 --> 00:14:08.040 align:middle line:84% like the tilt aftereffect or spatial frequency 00:14:08.040 --> 00:14:11.360 align:middle line:84% adaptation and the effect on the contrast sensitivity function. 00:14:11.360 --> 00:14:13.040 align:middle line:84% So there are perceptual phenomena 00:14:13.040 --> 00:14:15.200 align:middle line:84% that fall under the rubric of early vision. 00:14:15.200 --> 00:14:18.400 align:middle line:84% They often are used to make inferences 00:14:18.400 --> 00:14:22.293 align:middle line:84% about the mechanisms of early vision. 00:14:22.293 --> 00:14:26.060 align:middle line:84% So mid-level vision, by contrast, 00:14:26.060 --> 00:14:28.280 align:middle line:84% typically refers to processing stages 00:14:28.280 --> 00:14:31.220 align:middle line:84% that involve inferences that, in some way, 00:14:31.220 --> 00:14:34.480 align:middle line:84% are about the world, that are based on measurements that 00:14:34.480 --> 00:14:38.062 align:middle line:84% are made in early vision, eventually leading up 00:14:38.062 --> 00:14:40.520 align:middle line:84% to object recognition and scene perception, which typically 00:14:40.520 --> 00:14:42.420 align:middle line:84% people would call high-level vision. 00:14:42.420 --> 00:14:44.803 align:middle line:84% So we got early vision, mid-level vision, 00:14:44.803 --> 00:14:45.720 align:middle line:90% and high-level vision. 00:14:45.720 --> 00:14:47.760 align:middle line:84% And these are loose and sloppy terms, 00:14:47.760 --> 00:14:49.260 align:middle line:84% but they're nonetheless kind of used 00:14:49.260 --> 00:14:52.340 align:middle line:84% to indicate different parts of the field 00:14:52.340 --> 00:14:55.420 align:middle line:84% and different aspects of visual perception. 00:14:55.420 --> 00:14:57.220 align:middle line:84% And the other thing that I should say 00:14:57.220 --> 00:15:01.260 align:middle line:84% is that unlike early vision, where there's 00:15:01.260 --> 00:15:05.980 align:middle line:84% often fairly reasonable linkages between some 00:15:05.980 --> 00:15:10.140 align:middle line:84% of the associated perceptual phenomena and neurons, typically 00:15:10.140 --> 00:15:13.400 align:middle line:84% retina, LGN, and V1, mid-level vision 00:15:13.400 --> 00:15:15.340 align:middle line:84% and the various perceptual phenomena that 00:15:15.340 --> 00:15:17.540 align:middle line:84% are associated with it is less well 00:15:17.540 --> 00:15:20.780 align:middle line:84% linked to specific anatomical stages of the visual system 00:15:20.780 --> 00:15:21.960 align:middle line:90% and to individual neurons. 00:15:21.960 --> 00:15:23.960 align:middle line:84% And so in some cases, those relationships exist, 00:15:23.960 --> 00:15:28.540 align:middle line:84% and we'll talk about them, but they're not as tight. 00:15:28.540 --> 00:15:34.500 align:middle line:84% So one important theme that's central to mid-level vision 00:15:34.500 --> 00:15:38.820 align:middle line:84% is the idea that local measurements are ambiguous. 00:15:38.820 --> 00:15:43.280 align:middle line:84% So neurons in the visual system in some sense 00:15:43.280 --> 00:15:46.460 align:middle line:84% view the world through these little apertures, right? 00:15:46.460 --> 00:15:48.240 align:middle line:90% They have receptive fields. 00:15:48.240 --> 00:15:51.240 align:middle line:84% So that means that there's a region of visual space 00:15:51.240 --> 00:15:53.640 align:middle line:90% that drives their response. 00:15:53.640 --> 00:15:56.600 align:middle line:84% So they're looking at the world through these localized spatial 00:15:56.600 --> 00:15:59.360 align:middle line:84% receptive fields, measuring something 00:15:59.360 --> 00:16:02.880 align:middle line:84% that happens here or happens here or happens here. 00:16:02.880 --> 00:16:08.160 align:middle line:84% And these local measurements are often quite ambiguous. 00:16:08.160 --> 00:16:11.200 align:middle line:84% So they can be ambiguous in terms of what is actually 00:16:11.200 --> 00:16:15.240 align:middle line:84% causing the image intensities that's 00:16:15.240 --> 00:16:17.220 align:middle line:84% within that local region that's being measured. 00:16:17.220 --> 00:16:23.000 align:middle line:84% So this is an image that was designed 00:16:23.000 --> 00:16:25.920 align:middle line:90% by my PhD advisor, Ted Adelson. 00:16:25.920 --> 00:16:29.800 align:middle line:84% And it depicts a simple thing that 00:16:29.800 --> 00:16:32.400 align:middle line:84% looks like it's painted different colors. 00:16:32.400 --> 00:16:34.620 align:middle line:84% And so you have these different kinds of edges. 00:16:34.620 --> 00:16:37.480 align:middle line:84% So you have this edge here and this edge here. 00:16:37.480 --> 00:16:39.780 align:middle line:84% And inside those apertures, the image 00:16:39.780 --> 00:16:41.940 align:middle line:84% is exactly the same in the two cases. 00:16:41.940 --> 00:16:45.400 align:middle line:84% There's light gray on one side and dark gray on the other side. 00:16:45.400 --> 00:16:48.540 align:middle line:84% But you can just tell from looking at it that this 00:16:48.540 --> 00:16:50.820 align:middle line:90% is a change in pigmentation. 00:16:50.820 --> 00:16:53.340 align:middle line:84% It looks like maybe a change in the color of paint 00:16:53.340 --> 00:16:54.760 align:middle line:90% that was used to paint it. 00:16:54.760 --> 00:16:58.058 align:middle line:84% Whereas this is due to what's called shading. 00:16:58.058 --> 00:17:00.100 align:middle line:84% The fact that there's different image intensities 00:17:00.100 --> 00:17:04.660 align:middle line:84% there is because the surface orientation changes. 00:17:04.660 --> 00:17:06.579 align:middle line:84% So of course, you look at this thing, 00:17:06.579 --> 00:17:09.819 align:middle line:84% and your visual system, acting in concert, 00:17:09.819 --> 00:17:12.440 align:middle line:84% is able to come up with the correct interpretation. 00:17:12.440 --> 00:17:14.740 align:middle line:84% But the point is that the local measurements 00:17:14.740 --> 00:17:17.140 align:middle line:84% that are being made early in your visual system 00:17:17.140 --> 00:17:18.300 align:middle line:90% are ambiguous. 00:17:18.300 --> 00:17:20.619 align:middle line:84% And an individual local measurement on its own 00:17:20.619 --> 00:17:23.619 align:middle line:84% is not enough to tell you what's actually happening in the world 00:17:23.619 --> 00:17:28.300 align:middle line:84% to cause that particular pattern. 00:17:28.300 --> 00:17:33.340 align:middle line:84% You can see similar kinds of things in actual images. 00:17:33.340 --> 00:17:36.020 align:middle line:84% So these are some interesting cases where-- 00:17:36.020 --> 00:17:38.340 align:middle line:84% so each of these patches here-- this one, 00:17:38.340 --> 00:17:39.600 align:middle line:90% this one and this one-- 00:17:39.600 --> 00:17:46.080 align:middle line:84% are kind of close-ups of three regions of the edge of this log. 00:17:46.080 --> 00:17:49.660 align:middle line:84% So there's just a log on a background of stones. 00:17:49.660 --> 00:17:51.700 align:middle line:84% And again, when you look at the image, 00:17:51.700 --> 00:17:54.220 align:middle line:84% it's obvious that there's the edge of a log there. 00:17:54.220 --> 00:17:58.200 align:middle line:84% But then, if you zoom in to what's actually 00:17:58.200 --> 00:18:01.520 align:middle line:84% evident at the level of a receptive field, 00:18:01.520 --> 00:18:04.660 align:middle line:84% you can see that it's really not very clear. 00:18:04.660 --> 00:18:06.380 align:middle line:84% So in particular, like this case, 00:18:06.380 --> 00:18:09.920 align:middle line:84% for instance, if you look closely, the edge of the log 00:18:09.920 --> 00:18:11.760 align:middle line:90% is kind of right here. 00:18:11.760 --> 00:18:15.760 align:middle line:84% But in fact, all the contrast in this particular local region 00:18:15.760 --> 00:18:17.640 align:middle line:84% is up here because there's a shadow being 00:18:17.640 --> 00:18:21.880 align:middle line:84% cast by a rock that's right next to that edge. 00:18:21.880 --> 00:18:25.625 align:middle line:84% And so the point being that if all you had to analyze was this, 00:18:25.625 --> 00:18:27.000 align:middle line:84% and you were trying to figure out 00:18:27.000 --> 00:18:31.880 align:middle line:84% where the object boundary was, it'd be pretty hard to do. 00:18:31.880 --> 00:18:36.410 align:middle line:84% And the regions vary in terms of how much evidence there 00:18:36.410 --> 00:18:40.330 align:middle line:84% is for that edge, but they all have some degree of ambiguity. 00:18:40.330 --> 00:18:45.730 align:middle line:84% And that's actually a really useful exercise to do. 00:18:45.730 --> 00:18:50.850 align:middle line:84% If you really want to appreciate this more, 00:18:50.850 --> 00:18:54.210 align:middle line:84% you can take some images and view them 00:18:54.210 --> 00:18:55.610 align:middle line:90% through a little aperture. 00:18:55.610 --> 00:18:57.330 align:middle line:84% And if you look around with the aperture, 00:18:57.330 --> 00:18:59.550 align:middle line:84% it's really hard to make sense of what's going on. 00:18:59.550 --> 00:19:03.490 align:middle line:90% 00:19:03.490 --> 00:19:06.970 align:middle line:84% So local measurements are ambiguous. 00:19:06.970 --> 00:19:09.410 align:middle line:84% In order to make inferences about the world, 00:19:09.410 --> 00:19:12.370 align:middle line:84% somehow they have to be combined in some way. 00:19:12.370 --> 00:19:14.730 align:middle line:84% And so one kind of phenomenon that 00:19:14.730 --> 00:19:19.450 align:middle line:84% has some relationship to that general idea 00:19:19.450 --> 00:19:21.570 align:middle line:84% and is typically associated with mid-level vision 00:19:21.570 --> 00:19:23.090 align:middle line:90% is perceptual grouping. 00:19:23.090 --> 00:19:27.850 align:middle line:84% So this refers to the fact that things that are similar 00:19:27.850 --> 00:19:31.370 align:middle line:84% tend to subjectively group together, 00:19:31.370 --> 00:19:35.670 align:middle line:84% in the sense that they appear to be part of the same thing. 00:19:35.670 --> 00:19:39.070 align:middle line:84% And there was a movement in psychology a long time 00:19:39.070 --> 00:19:40.467 align:middle line:90% ago called gestalt psychology. 00:19:40.467 --> 00:19:42.550 align:middle line:84% And one of the main things they were interested in 00:19:42.550 --> 00:19:44.970 align:middle line:90% were these grouping rules. 00:19:44.970 --> 00:19:46.170 align:middle line:90% So what is this all about? 00:19:46.170 --> 00:19:51.870 align:middle line:84% So if you look at these two images, 00:19:51.870 --> 00:19:54.710 align:middle line:84% there's a sense in which you see these images 00:19:54.710 --> 00:19:57.470 align:middle line:90% as consisting of rows. 00:19:57.470 --> 00:19:59.750 align:middle line:84% You look at them, and the natural description 00:19:59.750 --> 00:20:01.990 align:middle line:84% is that you have rows of circles and squares 00:20:01.990 --> 00:20:03.790 align:middle line:90% or black dots and green dots. 00:20:03.790 --> 00:20:08.270 align:middle line:84% And so that subjective sense, that these circles kind of 00:20:08.270 --> 00:20:10.710 align:middle line:84% belong together and these circles belong together, 00:20:10.710 --> 00:20:11.970 align:middle line:90% is called grouping. 00:20:11.970 --> 00:20:15.510 align:middle line:90% 00:20:15.510 --> 00:20:19.110 align:middle line:84% And in general, similarity is a major factor that kind of 00:20:19.110 --> 00:20:20.290 align:middle line:90% determines grouping. 00:20:20.290 --> 00:20:22.910 align:middle line:90% 00:20:22.910 --> 00:20:26.110 align:middle line:84% Common fate-- so the fact that things move together-- that 00:20:26.110 --> 00:20:28.550 align:middle line:90% has a really big effect. 00:20:28.550 --> 00:20:31.350 align:middle line:84% So that thing that kind of looks like a number 4, 00:20:31.350 --> 00:20:34.250 align:middle line:84% they all look like they're part of the same thing. 00:20:34.250 --> 00:20:37.475 align:middle line:90% There's texture grouping. 00:20:37.475 --> 00:20:39.550 align:middle line:90% There's grouping by proximity. 00:20:39.550 --> 00:20:42.570 align:middle line:84% So again, you look at the thing on the left, 00:20:42.570 --> 00:20:45.433 align:middle line:84% and you see these horizontal rows, 00:20:45.433 --> 00:20:47.850 align:middle line:84% presumably because the circles are kind of closer together 00:20:47.850 --> 00:20:50.230 align:middle line:84% in the horizontal dimension than the vertical dimension. 00:20:50.230 --> 00:20:55.150 align:middle line:84% You look here, and you tend to see this as two groups of dots. 00:20:55.150 --> 00:20:59.170 align:middle line:90% 00:20:59.170 --> 00:20:59.670 align:middle line:90% All right. 00:20:59.670 --> 00:21:01.045 align:middle line:84% So there's lots of different ways 00:21:01.045 --> 00:21:02.870 align:middle line:84% to think about these sorts of phenomena. 00:21:02.870 --> 00:21:05.537 align:middle line:84% And you remember this idea that there are these different levels 00:21:05.537 --> 00:21:06.308 align:middle line:90% of analysis. 00:21:06.308 --> 00:21:07.850 align:middle line:84% There's the computational level where 00:21:07.850 --> 00:21:10.330 align:middle line:84% we can talk about the problem that's being solved 00:21:10.330 --> 00:21:13.170 align:middle line:84% and the constraints that allow it to be solved. 00:21:13.170 --> 00:21:14.547 align:middle line:90% There's the algorithmic level. 00:21:14.547 --> 00:21:16.130 align:middle line:84% There's the implementation level where 00:21:16.130 --> 00:21:17.963 align:middle line:84% you talk about how you would describe things 00:21:17.963 --> 00:21:20.050 align:middle line:90% in terms of neural circuitry. 00:21:20.050 --> 00:21:22.570 align:middle line:84% And it's possible to think about grouping at all 00:21:22.570 --> 00:21:24.610 align:middle line:90% of these different levels. 00:21:24.610 --> 00:21:28.490 align:middle line:84% So for instance, we can give an explanation of grouping 00:21:28.490 --> 00:21:31.050 align:middle line:84% in terms of neurons, for instance, 00:21:31.050 --> 00:21:34.270 align:middle line:84% by saying that there are neurons in the brain that respond more 00:21:34.270 --> 00:21:36.750 align:middle line:84% strongly when their neighbors, which 00:21:36.750 --> 00:21:39.690 align:middle line:84% will have nearby receptive fields, are also responding. 00:21:39.690 --> 00:21:41.630 align:middle line:84% And in fact, we'll see some evidence 00:21:41.630 --> 00:21:43.950 align:middle line:84% that there's probably some truth to that. 00:21:43.950 --> 00:21:47.230 align:middle line:84% And so you get a strong response when many dots are in a line 00:21:47.230 --> 00:21:48.670 align:middle line:90% or in a clump. 00:21:48.670 --> 00:21:54.310 align:middle line:84% So that's one way to explain how the perceptual effect 00:21:54.310 --> 00:21:56.630 align:middle line:84% of grouping would come about from neurons, 00:21:56.630 --> 00:21:58.930 align:middle line:84% but it doesn't really tell us why that would happen. 00:21:58.930 --> 00:22:05.070 align:middle line:84% So alternatively, we could talk about grouping 00:22:05.070 --> 00:22:06.450 align:middle line:90% in terms of probability. 00:22:06.450 --> 00:22:09.350 align:middle line:84% So the Helmholtzian approach to thinking 00:22:09.350 --> 00:22:11.870 align:middle line:84% about this at the computational level 00:22:11.870 --> 00:22:13.670 align:middle line:84% would be the idea that what we see 00:22:13.670 --> 00:22:16.870 align:middle line:84% is our best guess as to what is in the world based on the input 00:22:16.870 --> 00:22:19.150 align:middle line:84% data and based on our prior experience. 00:22:19.150 --> 00:22:20.790 align:middle line:84% So you might suppose that when things 00:22:20.790 --> 00:22:22.610 align:middle line:84% are close to each other in the image, 00:22:22.610 --> 00:22:24.402 align:middle line:84% there's a good chance that they're actually 00:22:24.402 --> 00:22:26.382 align:middle line:84% part of the same object in the world. 00:22:26.382 --> 00:22:29.090 align:middle line:84% And so we have a tendency to see them as part of the same object, 00:22:29.090 --> 00:22:32.440 align:middle line:84% because that sense that we see them 00:22:32.440 --> 00:22:35.880 align:middle line:84% as the same thing represents an inference that they're actually 00:22:35.880 --> 00:22:39.820 align:middle line:84% caused by a single object in the world. 00:22:39.820 --> 00:22:41.320 align:middle line:84% So the idea here is that there are 00:22:41.320 --> 00:22:45.320 align:middle line:84% these heuristics that are based on probabilities in the world. 00:22:45.320 --> 00:22:48.152 align:middle line:84% And that's a perfectly useful way 00:22:48.152 --> 00:22:50.360 align:middle line:84% to think about this phenomena, but it doesn't tell us 00:22:50.360 --> 00:22:53.040 align:middle line:84% anything about how you would implement this with neurons. 00:22:53.040 --> 00:22:54.680 align:middle line:90% Do you have a question? 00:22:54.680 --> 00:22:56.080 align:middle line:90% STUDENT: Yeah. 00:22:56.080 --> 00:23:01.040 align:middle line:84% So I saw a poster outside your office, something 00:23:01.040 --> 00:23:04.160 align:middle line:84% along the lines of testing these illusions 00:23:04.160 --> 00:23:12.680 align:middle line:84% with legally blind kids who just had surgery and gotten 00:23:12.680 --> 00:23:15.360 align:middle line:90% their vision back. 00:23:15.360 --> 00:23:20.720 align:middle line:84% So if this Helmholtz explanation actually is true, 00:23:20.720 --> 00:23:23.320 align:middle line:90% does this not work? 00:23:23.320 --> 00:23:26.400 align:middle line:84% JOSH MCDERMOTT: Does what not work? 00:23:26.400 --> 00:23:29.460 align:middle line:84% STUDENT: Would a person that just 00:23:29.460 --> 00:23:36.100 align:middle line:84% got their vision from a surgery see the right side, for example? 00:23:36.100 --> 00:23:37.460 align:middle line:90% JOSH MCDERMOTT: Yeah. 00:23:37.460 --> 00:23:40.060 align:middle line:84% So the question is, would someone 00:23:40.060 --> 00:23:44.100 align:middle line:84% who's been visually impaired or blind from birth, who suddenly 00:23:44.100 --> 00:23:48.640 align:middle line:84% has their vision restored, would these effects hold for them? 00:23:48.640 --> 00:23:51.535 align:middle line:90% And I don't know. 00:23:51.535 --> 00:23:53.160 align:middle line:84% I mean, something that could be tested, 00:23:53.160 --> 00:23:55.900 align:middle line:84% but I think it really just depends 00:23:55.900 --> 00:24:01.820 align:middle line:84% on the extent to which all these factors are really 00:24:01.820 --> 00:24:05.060 align:middle line:84% something that result from evolution or result 00:24:05.060 --> 00:24:06.732 align:middle line:90% from development. 00:24:06.732 --> 00:24:08.440 align:middle line:84% And it could be some combination of both. 00:24:08.440 --> 00:24:12.900 align:middle line:84% So one possibility is that your brain is wired up 00:24:12.900 --> 00:24:15.820 align:middle line:84% in a certain way in order to leverage 00:24:15.820 --> 00:24:17.780 align:middle line:84% these probabilistic relationships that 00:24:17.780 --> 00:24:19.180 align:middle line:90% exist in the world. 00:24:19.180 --> 00:24:22.820 align:middle line:84% So natural selection caused the brain 00:24:22.820 --> 00:24:27.900 align:middle line:84% to evolve to give rise to this kind of grouping. 00:24:27.900 --> 00:24:28.820 align:middle line:90% That's possible. 00:24:28.820 --> 00:24:31.320 align:middle line:84% And the prediction there would be that, well, from birth, 00:24:31.320 --> 00:24:34.980 align:middle line:84% or maybe largely independent of developmental experience, 00:24:34.980 --> 00:24:37.540 align:middle line:84% you would experience this more or less the same. 00:24:37.540 --> 00:24:40.240 align:middle line:84% But the other possibility is that these 00:24:40.240 --> 00:24:44.660 align:middle line:84% are things that we learn online over the course of development, 00:24:44.660 --> 00:24:47.120 align:middle line:84% just by our interactions with the world, 00:24:47.120 --> 00:24:51.360 align:middle line:84% by looking out and then grabbing things and realizing 00:24:51.360 --> 00:24:53.380 align:middle line:84% what actually constitutes an object and stuff. 00:24:53.380 --> 00:25:01.040 align:middle line:84% And in general, we typically don't 00:25:01.040 --> 00:25:04.120 align:middle line:84% know a whole lot about the relative importance 00:25:04.120 --> 00:25:07.020 align:middle line:84% of development and evolution for a lot of things in perception. 00:25:07.020 --> 00:25:11.100 align:middle line:84% So a lot of those questions are still not very well answered. 00:25:11.100 --> 00:25:14.240 align:middle line:84% And so in the case of grouping, I don't really 00:25:14.240 --> 00:25:15.263 align:middle line:90% know the answer to that. 00:25:15.263 --> 00:25:16.680 align:middle line:84% And another way to test that would 00:25:16.680 --> 00:25:19.720 align:middle line:84% be to ask, well, what would happen 00:25:19.720 --> 00:25:22.080 align:middle line:84% if somebody was reared in a world that 00:25:22.080 --> 00:25:24.480 align:middle line:90% had very different statistics? 00:25:24.480 --> 00:25:27.020 align:middle line:84% Would they end up having very different perception? 00:25:27.020 --> 00:25:30.230 align:middle line:84% That's another way you could, in principle, ask that question. 00:25:30.230 --> 00:25:33.410 align:middle line:84% Of course, that natural experiment 00:25:33.410 --> 00:25:35.530 align:middle line:90% doesn't typically occur. 00:25:35.530 --> 00:25:38.190 align:middle line:84% And there's no way to do that in people often. 00:25:38.190 --> 00:25:40.010 align:middle line:84% But sometimes you can do those experiments 00:25:40.010 --> 00:25:41.930 align:middle line:90% with animals in the lab. 00:25:41.930 --> 00:25:44.210 align:middle line:84% And you might in principle be able to do that. 00:25:44.210 --> 00:25:45.370 align:middle line:90% OK. 00:25:45.370 --> 00:25:45.990 align:middle line:90% OK. 00:25:45.990 --> 00:25:47.810 align:middle line:84% So this is perceptual grouping, and there's 00:25:47.810 --> 00:25:50.170 align:middle line:84% these different levels of explanation. 00:25:50.170 --> 00:25:52.575 align:middle line:84% And there's lots of factors that affect grouping. 00:25:52.575 --> 00:25:54.950 align:middle line:84% So this is what's known as grouping by good continuation. 00:25:54.950 --> 00:25:57.250 align:middle line:90% So you see an image like this. 00:25:57.250 --> 00:26:00.930 align:middle line:84% And what you perceive is a circle and a square. 00:26:00.930 --> 00:26:03.610 align:middle line:84% Now this is a perfectly valid description 00:26:03.610 --> 00:26:05.070 align:middle line:84% of how you would create that image. 00:26:05.070 --> 00:26:06.970 align:middle line:84% You got a Pac-Man and then this other funny shape 00:26:06.970 --> 00:26:08.928 align:middle line:84% that are just placed together in a certain way. 00:26:08.928 --> 00:26:10.450 align:middle line:90% But that's not what you see. 00:26:10.450 --> 00:26:12.650 align:middle line:84% So what you tend to see are things 00:26:12.650 --> 00:26:15.330 align:middle line:84% that are kind of continuous contours. 00:26:15.330 --> 00:26:17.210 align:middle line:90% You also group by closure. 00:26:17.210 --> 00:26:19.930 align:middle line:84% So this image on the right could be 00:26:19.930 --> 00:26:21.870 align:middle line:84% described in a whole bunch of different ways. 00:26:21.870 --> 00:26:25.470 align:middle line:84% For instance, you could see it mostly as these closed shapes. 00:26:25.470 --> 00:26:28.123 align:middle line:84% You could see it as these kind of cross-like shapes. 00:26:28.123 --> 00:26:28.790 align:middle line:90% You can see it-- 00:26:28.790 --> 00:26:30.430 align:middle line:84% I don't what that-- what would you call 00:26:30.430 --> 00:26:32.530 align:middle line:90% that, a funny x-like thing? 00:26:32.530 --> 00:26:35.670 align:middle line:84% They're all valid descriptions of the image. 00:26:35.670 --> 00:26:38.210 align:middle line:84% But people tend to see the circular organization. 00:26:38.210 --> 00:26:41.150 align:middle line:84% And again, then the naive idea is 00:26:41.150 --> 00:26:46.030 align:middle line:84% that, well, closed contours are pretty common in the world. 00:26:46.030 --> 00:26:47.735 align:middle line:90% Objects tend to be like that. 00:26:47.735 --> 00:26:49.110 align:middle line:84% And so your visual system kind of 00:26:49.110 --> 00:26:51.510 align:middle line:84% is wired up or programmed in some way 00:26:51.510 --> 00:26:54.190 align:middle line:90% to enable you to see that. 00:26:54.190 --> 00:26:54.737 align:middle line:90% OK. 00:26:54.737 --> 00:26:55.570 align:middle line:90% I want to end there. 00:26:55.570 --> 00:27:00.950 align:middle line:84% When we come back, we will resume talking about grouping. 00:27:00.950 --> 00:27:03.190 align:middle line:90% Have a good weekend. 00:27:03.190 --> 00:27:04.830 align:middle line:84% And actually, it's spring break, right? 00:27:04.830 --> 00:27:07.270 align:middle line:84% So I hope everybody has a great spring 00:27:07.270 --> 00:27:10.710 align:middle line:84% break, that you get a chance to kick back a little bit. 00:27:10.710 --> 00:27:13.710 align:middle line:84% And I will see you in, like, 10 days. 00:27:13.710 --> 00:27:15.260 align:middle line:90% Bye. 00:27:15.260 --> 00:27:26.000 align:middle line:90%