 Just as going live still. Okay, so it looks like we're live. Excellent, okay, so hi everyone and welcome to another Sussex Vision Talk. My name is Tessa Herzog and I'm a PhD student in the labs of Tom Barden and Leon Lanyardo. I'm really pleased to host today's seminar which is a collaboration between the highly successful Worldwide Neuro Talks and the Sussex Libi-Hume Seminar Series. The latter is run by the Libi-Hume doctoral students here at Sussex University where we host speakers exploring topics from sensation to perception and awareness. You can check the schedule for the upcoming talks for both the WWN Talks and the Libi-Hume Seminar Series at the links in the video description below. So today I'm very happy to welcome Dr Bevel Conway who's going to be talking about the neuroscience of colour. Dr Conway is a senior investigator at the National Eye Institute in the laboratory of sensory motor research. So his lab aims to understand the brain processes by which sensory data are transformed into perceptions, thoughts and actions. The work in his lab has been especially focused on developing colour as a model system and the lab uses a combination of techniques including psychophysics and non-invasive brain imaging including MRI and MEG in humans alongside experimental fMRI guided microelectrode recording, fMRI guided pharmacological blockade and computational modeling. Dr Conway is also a visual artist and he's written on the intersection between art practice and neuroscience. So I'm really happy to hand over to you Bevel so you can start screen sharing. Great thanks so much here we go let's see if this works okay and now all right which screen do you guys see we've got your notes screen still okay and yeah there we are yeah are we good all righty okay thanks so much um it's a super wonderful privilege to be here I'm just gonna start a clock here because I don't want to try and keep on time um there we are all right so um yeah thanks so much for the invitation Tessa it's nice to meet you and I'm looking forward to a discussion afterwards before I launch in I do want to just acknowledge um the folks in my lab over the last few years who've contributed to the work I'll describe I'll point out specific contributions throughout my talk um so a broad goal of visual neuroscience is to understand how the brain generates knowledge and as you heard my labs especially interested in using color to approach this question and I admit that it might seem on the surface to be a bit silly because it's widely assumed that color does not really require much brain processing as you guys all know in Sussex fish do it pretty well um and it um and in any event it's often thought to just be a low-level stimulus feature of limited utility division uh so why would we use this as a model system um it has been surprisingly difficult to quantitatively determine the benefit of of color division uh so I'm gonna go through a little background to try and motivate why why we're doing this why we use color in this way okay so in a famous study Beederman and June showed that for most objects color doesn't afford much recognition benefit if anything color uh recognition error rates actually higher for colored images than for line drawings uh Gagan Fertner and Rieger pushed back against this idea that uh color was completely useless they used an image recognition task and showed that um if both the test and the probe images are colored that um you recognize and you get a slight recognition benefit so that this supports the idea that that color helps you recognize things faster and remember them better but as you'll see from those graphs the benefit is really modest and it's not clear that this benefit actually relates to object vision per se it could be a kind of an intentional mechanism indeed for most objects such as faces colors neither sufficient nor necessary by itself for recognition so it's very hard in this uh image to tell the identity of the face because it's an equiluminate picture there's no luminance contrast in fact luminance cues are almost entirely sufficient to recognize most objects and this is really why computational models of object vision such as the original h max very influential algorithm don't incorporate color because you just don't really need color to recognize objects now there are some data on the fringes color can aid face recognition under but under very degraded image conditions so this benefit is conferred for example by enhancing feature segmentation making the eyes and hairline more visible and not because the specific color itself is diagnostic of the particular face so as yippen sinna pointed out in this paper from 2002 you get the same recognition benefit uh in a false colored image um and again this benefit is pretty modest so most people can tell that that's lady diana even in the black and white picture but with the color either of the colored versions you get a tiny little boost okay so the upshot is that color doesn't really seem to be critical for object recognition but instead seems to support a range of behaviors of indirect relevance to object vision these include low level visual abilities such as scene segmentation so being able to parse objects from backgrounds using color cues and high level processes such as attentional capture grouping categorization memory communication emotion and reward and social cognition so it's a vast array of behaviors and i think this array of behaviors expands the potential objectives of vision that is when we think what is vision for we often go to object recognition and i think color sort of blows open that uh that range of possibilities and says color does a lot more and for me at least i i think these these objectives of vision are at least as important as object recognition itself the problem is we have very little idea how these and other behaviors of vision uh are enabled by the brain uh and give rise to knowledge okay so to make this overarching question tractable how does the brain generate knowledge i think it might be useful to sort of take a step back and break the question down into its two component parts make them explicit first what are the computational objectives of vision let's not take it for granted that we know what vision is for let's try and spell it out and then let let's figure out how those objectives are implemented in the brain when we have an idea about what vision is for i think it's a lot easier to hunt for neural implementations but as i hope to show you today the reverse is also true sometimes uh the quest to understand how the brain works can prompt a reevaluation of ideas about what vision is for okay let's dive in so a few years ago uh we ran some imaging experiments in monkeys to test alternative hypotheses about the organization of the visual brain this is a side view of the macaque brain that has been computationally inflated to reveal the cortical sheet that would otherwise be buried in the cell seat now a considerable amount of the macaque cerebral cortex is implicated in vision indicated by this dashed ellipse the tissue towards the back can be carved up by the meridian representations into the classic retinotopic areas of v1 v2 v3 and v4 but there's this large swath of tissue anterior to v4 the infrotemporal cortex whose organization is poorly understood and we were interested in trying to figure out how that bit of brain is organized there are two proposals for it organization at least uh among the set are the swiss army knife proposal championed by nancy canwisher that calls for specialized modules embedded in multipurpose tissue so you might have an island of face processing tissue inside this tissue that's more general purpose and the distributed concept processing model uh championed by uh jim haxby which argued that basically all parts of it are contributing more or less to to most of object vision so a few years ago rosa lay for susan i tested these ideas by mapping the relative activation patterns in the same individual animals to a battery of stimuli and we found a large network of color bias domains in it so uh they're shown in the cool colors in this lateral view from the cac brain what's more these regions these color bias domains were systematically positioned next to face patches shown in warm colors uh the activation patterns seem to carve up it into four sub areas each defined by its own complete set of functional activation patterns so these and other data we obtain suggests that infrotemporal cortex is organized according to parallel multi-stage processing streams for faces colors places and objects and perhaps other stimulus attributes so instead of just being a big mush it's actually got these rather discrete stages to it now the hypothesis is supported by anatomical connectivity data in a large meta analysis by cravitz et al of track tracing data they discovered that the predominant pattern of connectivity in it is parallel connections running along the posterior anterior axis and they identified these four main nodes teod te pd te ad and tgv what's super cool is these anatomical nodes align with the functional domains suggestive of a beautiful form functional association that the anatomy is supporting this functional architecture that we discovered um so taken together these results suggest that the face patches are not some unique weird set of islands of very specialized tissue but they're instead one manifestation of a canonical set of computations performed in parallel streams along it and i think that's a really powerful way of thinking about it and it reconciles these two theories i introduced earlier borrowing key elements of both the distributed model and the specialized model the the the framework is appealing not only because it matches the empirical data but also because it comes with a hypothesis for how it develops and how it evolved which became apparent except in experiments we did to try and figure out the underlying organizing principle and i'll talk about those in a second i think it's also powerful when they're converging lines of evidence and there are new papers from uh sri hasam et al from nature neuroscience and baudel and nature that really support this this architectural model this parallel multi-stage framework model okay so what's the underlying organizing principle that determines this structure well a key clue came from rafael malik's group who found a coarse eccentricity bias across higher order object areas in humans now we ran experiments to determine whether monkeys have a similar eccentricity bias they did and in a very precise pattern so besides the foveal representation in early areas just i outlined it there with that big star and in area mt that's that dorsal area there there were a sequence of foveal representations along it there are three here one two and three and these representations line up with the functional domains discovered using colors and faces so you can see them on the other hemisphere there so that suggests that it is organized by repeated eccentricity templates perhaps inherited from uh v1 recent work from arcaro and livingstone suggests that this proto eccentricity template structure we discovered is present in primates from birth so the eccentricity template results recall alman and casus reminder that a common mechanism of evolution is the replication of body parts so we see this uh in lots of domain the vertebral column or the segments in drosophila or the genetics of color vision the replication hypothesis has also been used to explain the expansion of retinotopic areas so this is in alman and casus framework how we got v1 v2 v3 and v4 all with a shared foveal and peripheral representation so our discovery of a series of eccentricity maps down it suggests that the theory can be expanded so according to this hypothesis an ancestral v4 duplicated to create primitive it which itself duplicated several times carrying forward this eccentricity representation it inherits from v1 so the idea is that foveal biases within it took on a role in computing faces peripheral representations took on a role in computing scenes and para foveal regions in between took on a role in computing colors and i i think that this hypothesis is quirky and interesting for for lots of reasons but one of them is it sort of relates to this weird question what's the cognitive scale of color so if foveal representations are taking on a role in face processing that makes sense because face perception requires the ability to parse objects finer than the scale of the object so you got to look at the features within the object whereas the far peripheral regions taking on a role in computing scenes makes sense because scene perception requires us to take in the entire visual field the color bias regions are situated in between which suggests that color is really serving this role on the level of an object and and and we've done some informal work just polling people if you ask them what's the cognitive scale of colors people are like huh but if you show them a multicolored object and you say what's its color or what are the colors or how would you describe it people will typically pick one color that's the kind of handle that they've used to describe the whole thing yeah so sorry to interrupt you the right hand side of your screen is slightly cropped and it wasn't too much of a problem before but now it's interfering with some of the is that still crop there yes it is yeah i'm wondering if the format of your slides is slightly is it wide screen i don't know are you zoomed in not sure we can't we can continue it's just a very small portion in the right hand side okay on the right hand side all right well so does it does it go through all of the in terms of the evolutionary time arrow can you see the arrowhead no we can't how much of the a i t c i t can you see so in the fifth globe there we see up to the letters p i t okay yeah all right i'll keep that in mind hopefully we can have a go at changing the slides sizes now if you'd like or you can just press on it's up to you i'm not sure how to do that do you know um yeah i can all right let's try okay pause we get to pause now give everybody a chance it's just one it's gone full screen now so i'm hoping it's full screen for everybody else as well okay here is that all right yeah that's perfect that's great you can see there ahead all right thank you so much okay so um it's good to be forced to take a little break anyway okay so before we get too exotic you might be wondering if the relationship of face patches and color bias regions we see in macaques is also evident in humans as you might predict if the activation patterns reflect some kind of ancestral homology the answer is yes so in a collaboration with Nancy can we're shown Rosalie for Susa we showed that humans have the same tripartite organization of color bias domains sandwiched between face domains and place domains this was published in a paper in 2016 in J Neuroside so the results suggest that the macaque is a good model of the human case and they're also consistent with this idea that color plays an important role in high level vision recall it in fro temporal cortex this large swath of tissue and both the human and the monkey has been implicated in object recognition in fact we also found color bias domains in prefrontal cortex of the macaque so this is just stimulus driven these beautiful domains again adjacent to face patches in the frontal in the frontal cortex so take a step back there's like as much tissue about as much tissue involved in color processing in parts of the brain implicated in high level vision as there is involved in face recognition what's all this color bias tissue doing i mean for one thing it's a strongly suggests that color is not simply just a low level stimulus feature and it returns us back to this question what is color for so the boundary between retinotopic areas in it reflects a transition from retinotopic or retina centric organizational systems to a cognitive one in other words the there's a transition in the brain from the back to the front that goes from how retinal information is encoded to how visual information is used in behavior so it's this critical transition that turns light information into knowledge you then care about and can use so given the literature implicating in fro temporal cortex and object vision we wondered if there was some statistical pattern in the colors of the parts of scenes that people care about as distinguished simply from the statistics in the environment so my colleague at MIT the linguist Ted Gibson taught me that one way of determining what people care about as opposed to just the statistics in the environment is to ask them what they label so we wondered about specifically the statistical distribution of the parts of scenes that participants label as objects and surprisingly there were no data on the color statistics of objects to address this gap sieve at nassingham in my lab analyzed a large database of photographs labeled with label object labels and these panels show random pixels from objects and backgrounds you can see right away as striking results the objects tend to be warmer colored compared to the backgrounds this plot quantifies the results to orient you u prime and v prime are the dimensions of the standard chromaticity space so our results show that objects differ from backgrounds along the u prime dimension and animate objects differ from inanimate objects along the perpendicular v prime dimension so I find this result really striking because it suggests that the dimensions of color space itself u prime and v prime do not derive from how color is encoded but instead reflect the high level use to which we put our color system namely the detection of parts of scenes that are likely to be relevant objects versus backgrounds and animate versus inanimate isabel rosenthal uh in my lab related the object color statistics to the color tuning responses measured with fmri so in this lateral view region showing high correlation of fmri color responses and object colors are depicted in yellow and they form this band along it so the results suggest that color provides us with semantic information that is contributing to the knowledge about the stuff in the world and this idea that color reflects the youthfulness of visual information as opposed to just the scene statistics is supported by the color naming studies we did together with tid gibson's lab so here's the question how many guesses does it take you to identify in a color array the chip i've picked out if all you know is the word that i use to label the color so we both have the same array i pick a chip you don't i'm not pointing at it i use a word how many guesses does it take you the answer is an indication of the communication efficiency of our language and it turns out that across all languages there is a striking universal pattern so this tapestry shows color chips for each language arranged in a row and they're arranged from left to right according to the rank order of the communication efficiency of that color within the language and the languages themselves are ordered top to bottom according to the overall efficiency of their language so at the top you have western languages that have very high communication efficiency regarding color very sophisticated color color naming systems and at the bottom you have less efficient languages that have fewer consensus color terms in the population but you can see despite how fancy your color naming system is there's this universal pattern that is warm colors are communicated more effectively than cool colors across all languages and those statistics are actually related or that pattern is actually related correlated with the object color statistics so the results suggest that the extent to which color is useful drives the color naming patterns which helps resolve a long-standing debate in the color naming literature but for our purposes the results provide insight into the role that color plays in how we acquire knowledge color doesn't just tell you so much about the specific identity of what you're looking at but rather the likelihood that it's something you'll care about that it'll be an object okay so there seems to be elements of a story emerging that it uses color for object recognition by becoming especially sensitive to the color statistics of objects but there's a hitch to this tidy account the band of it indicated by the yellow arrow that is most highly correlated with the color statistics of objects that band does not coincide with the color bias domains they tend to be ventral in fact if anything the band in it encompasses face patches more so than color bias regions which is peculiar why because face recognition doesn't need color you can test yourself with the selection of grayscale photographs of some luminaries in the neuroscience of color you can easily tell them apart even if you've never met them the fmri results show that the color signals influence many if not all parts of intro temporal cortex so we've got those color bias domains they they were identified because they show and i've an overt bias to color over luminance contrast but you also have this band of tissue with a bias for the colors of objects that encompasses the face patches so color is too simple of a term it's entering it in lots of different ways so marien duicke in the lab and others in my groups a group of post backs confirmed the mri was not misleading by doing fmri guided recording of the face patches so they used a clever stimulus set that preserves the luminance contrast relationships of the faces while parametrically varying hue i can walk you through how you make the stimulus set but the upshot is that any given face in the stimulus set still has the rich luminance contrast that the face system needs in order to respond to faces but we can make green faces and red faces and so on and sure enough the population of faces on average was biased for warm colors so there's the distribution of the population cell response you can see it peaks to the warm colors and has dips to the cool colors for reference the dashed curve shows the distribution of face colors from sophie werger's large sample of spectral measurements of faces across multiple ethnic ethnicities so the color tuning data are consistent with an earlier report by edwards et al that just measured the responses of face selective cells to natural versus unnaturally colored faces and showed that it cells face cells are often respond a little bit better to the naturally colored faces the complete color tuning functions that we obtained allow us to compute the fissure information for the population so the fissure information shows what colors are most readily discriminated by the population so so can these face selective cells with their color tuning actually distinguish between different face colors well there's a dip in the fissure information that coincides with the peak of the distribution of colors of faces which shows that the population is not actually optimally tuned to discriminate the colors of faces so to summarize face cells do not seem to be wired up to discriminate face colors they're using that color information to do something else instead i suspect the color tuning of it and of the face cells reflects this other key organizing principle so several reports have argued that infrotemple cortex is organized by responses to animacy measured using shape cues so this is work by nasa laris et al Shah et al and compland caramatsa they've really argued quite strongly that it is organized by this response to animacy and there are a number of other papers as well so our results suggest that color could contribute to this organizational structure by playing a role in animacy judgments especially those related to faces this returns us back to this question what is all that color bias tissue doing they don't seem to be encoding the colors of objects at least not in a generic way so it returns us to the cycle of questions one behavioral piece of evidence in favor of the notion of of of the role of color and object recognition comes from that earlier study by dagan furtner and reger i mentioned earlier that color helps us recognize things faster and to remember them better and and a big piece of data in support of that idea is that color and shape are somehow inextricably bound and the evidence for this comes from this kind of a demonstration where where we might ask what is the color of a banana there's a number of studies now that suggests that the color you associate with a banana isn't just an association the the gray banana actually looks like it's tinged with yellow and this is taken as evidence that the color and shape are processed together as two cues to object identity now we decided to test for these effects for both inanimate and animate objects using very rich shape cues and the reason really was that these effects the so-called cognitive penetrance of color the yellow tinging of the gray banana is thought to depend on the richness of the shape cues so if you make the shape cues really really rich as as you know lots of shading and so on you get a better effect than you do if you just use a line drawing well the richest shape cues we could imagine can be can be sort of created by looking at the world under monochromatic sodium light low pressure sodium light so mariam hasantash and rosalie for susa asked people to match the appearance of real objects and real people while immersed in this room lit with low pressure sodium light the monochromatic sodium light it has sort of very narrow bandwidth and you can think of it as a kind of knockout experiment it doesn't it prevents the retina from actually encoding color and the question is what colors do you see under this very peculiar circumstance where everything is essentially monochromatic or kind of weirdly colorless so if memory modulates the colors of objects then when retinal color encoding is impaired or disabled then the color matches should reflect the normal colors for color diagnostic objects just like the the gray banana should appear tinged with yellow but so so and we made that prediction that this should be the case for for objects that have a diagnostic color but not for objects like toys that don't have a diagnostic color so we have a little selection of toys and then we had an orange fruit a strawberry and a tomato and you can see those at the bottom and those tapas that tapestry there shows the color matches that people made 20 different subjects made to all of that array of different objects or samples under white light and it recovers basically what you expect now in addition and we'll get to that in a minute we also had people color match skin samples of real-life human models a Caucasian woman and an African-American woman we had two of each and we'll get to those results in a minute but for now what we noticed right away was that the matches under the low pressure sodium light did not really distinguish between the toys and the the naturally colored objects they all appeared the same kind of washed out brownie yellow color that's really just a color reflecting the illuminant so we didn't find very compelling evidence for this cognitive penetrance of color on on objects but there was one exception and that was faces so of the two races tested African-Americans and Caucasians every single subject matched their faces with a striking green so it wasn't the natural color it was this paradoxical color the effect was abolished when the face context was masked showing that this really isn't just about something about the skin it's actually the face context that's creating this paradoxical memory color now we did a detailed analysis of the color matches to determine the extent to which they modulated the long versus middle or the red greenish axis versus the scone systems and i'll dive into that now so under these plots show the multi-dimensional scaling relating the first dimension of the of the representational dissimilarity matrix for the s component along the s along the y-axis and along the m axis is the lm component so under white light the skin of faces and non-faces so face skin is shown as black symbols and non-face skin is shown as open symbols they're fully overlapping they're indistinguishable under white light people make the same matches to face skin and non-face skin what separates the populations is the race so Caucasians are in the upper left and African Americans are under in on the lower right what's quirky is under the low pressure sodium light you see a separation and that separation is selective for the lm dimension so this is regardless of race what we're now pulling out is a distinction based solely on the extent to which that memory color is modulating the red green or lm axis so i think this provides really quite a deep clue about the role of color and behavior it doesn't tell us about identity that is race but rather about behavioral state now although Yip and Sina do not describe it i think their original observations or the figure from their paper provides a kind of clue and don't you think Diana looks sick in the false colored picture i mean it the most striking feature to me of this image is not that color is helping us recognize her but that that false color she just looks really really ill and in fact many of the subjects surprisingly more women than men that was the only sex difference we found spontaneously reported that the faces under low pressure sodium light looked sick so males and females both gave the same color matches they both dialed it in to show green but the females were much more likely to spontaneously report there was something wrong about the face so um these results suggest that there is some kind of prior that your that your brain has some very rich information about the normal colors of faces but we've already seen that these face cells although they're tuned for warm colors they're not actually capable of discriminating the colors of faces but the behavioral results the color matching data i just presented they argue that it's not about discriminating colors around the color circle of faces but rather colors across the lm axis that's what the face selective cells are wired up for and if we replot our face selective our color tuning data from the fmri guided micro electrode recording you can see that the cells on average have this ramp shaped tuning function that is they respond more to faces if they have higher l versus m content and these ramp shaped tuning functions are very similar to the ones that doris tau and finrich frywald and market livingstone have discovered in face cells for other face parameters like the eye position or the spacing of the eyes to the nose and so on and so it looks like this is one parameter that the face selective cells are sensitive to that gives them their tuning and taken together the results then suggest that the cells the face selective cells they're not discriminating the face colors but instead they're encoding a prior about just the lm component and that lm component is the dynamic component of faces it's the component that varies with hemoglobin or oxygen saturation that tells you about emotion and health and so on so we return to where we started what are the computational objectives that vision and how are they implemented so we haven't really addressed this question yet we've still stuck the color bias regions sitting ventral to the face patches what are they really doing and i think at this point we can go back to the proposition that what is the color of a banana now when i ask you what's the color of a banana we reflexively answer that it's yellow and it's widely believed that yellow is somehow inextricably tied to object shape so much so that the gray bananas are thought to appear tinged with yellow those are the memory color experiments of hunts and and and alconon and dagenford and i described earlier so i don't challenge those results they're empirical findings there are some quirky features of those results namely that they seem to those memory colors seem to be most strongly evident for objects that have colors that align with the daylight access with the natural colors of the illuminant and and so it seems like the memory color effects recovered in those experiments may actually relate more to uncertainty about the illuminant than they do to the the inextricable yoking of color and shape information when i ask you what's the color of a banana i think really what we're asking is this kind of implied question what's the color of the bananas that you care about and i i think that particular clause that you care about is especially useful or important because it tells us that um that the color isn't relating to the identity but rather to the behavioral state of the object so as a banana ages it undergoes a number of color changes and those color changes don't really tell you much about the change in the identity the identity is evident in all of them and in the grayscale object the changes in color tell you about the changing state of the of the banana and the extent to which the banana is likely to be meaningful to you so when we ask what color is the banana we're drawing on this implicit role of color as it relates to use that is the cognitive function of color which we all take for granted is to tell us about whether or not something is likely to be relevant and i think this then provides a good formal argument for why object recognition and color biased regions are decoupled in the brain they serve different functions so the shape biased regions such as face patches enable object recognition and the color biased regions constitute this trainable system that promotes the computational efficiency efficient by connecting objects with their likely relevance and if color and shape were inextricably tied to each other and yoked to each other then they couldn't serve these independent functions so to test the implementation of this kind of broad hypothesis i think it's useful to break the problem down into mechanisms that encode color which we hypothesize are implemented in retinotopic cortex and mechanisms that decode what color means which i suspect depends on array of areas in infrotemple cortex and prefrontal cortex the human lesion literature provides some tantalizing clues to this these decoding mechanisms and we can discuss that at the end of the talk in a minute if that's of interest but it's a larger topic okay so to sum up i've shown you a bunch of data and i want to try and stitch it all together so we we used color as a tool to try and uncover the essentially the organizational structure of infrotemple cortex and we discovered that there's this parallel multi-stage processing of colors faces places and objects down the length of it and those results suggest not only that it comprises the set of somewhat discrete stages but it also strongly suggests that the face patch system or any system within it isn't unique but one exemplar one example of a canonical set of operations that take place down it and i think that's really useful as a framework because it means that whatever we learn from one of these systems we can use it to try and understand or create hypotheses in the other system and in fact this is one of the things we're doing at the moment is trying to figure out you know for specific hypotheses of what these stages in it are doing we're trying to use the the full array of of of tools that we have at our disposal and information about face processing place processing and color color processing importantly i refer to these domains these color bias regions in infrotemple cortex by this cumbersome term color bias regions because i don't think they're they're involved specifically in processing color i think that is done by the encoding stages already by the time we get to the v4 complex so we don't we actually know very little about what these domains are doing and i think there's lots of very exciting work yet to be done to figure out how they work and how they afford this range of cognitive features that are behaviors that we use color for one loose idea is that maybe infrotemple cortex is actually parallel streams where you've got one stream or set of streams that are telling you what stuff is the identity of stuff out there and another stream that's computing whether or not you should care about that stuff maybe that's what the color bias domains are feeding into so then we also went through data showing that the organizational structure in the infrotemple cortex is predicted by the set of eccentricity templates that are inherited from v1 and that then predict the organization so that the foveal bias for regions take on a role in playing in processing faces and peripheral regions take on a role in processing places and mid peripheral regions take on a role in processing colors and that quirky question of what's the cognitive scale of of color might relate to objects which i think is interesting because all of the work i've described about color is about objects at a graspable scale that is we use color to identify that which we can then touch or interact with and that may be partly why we don't or languages typically don't develop color terms for blue until relatively late in the languages evolution because they're despite the fact that the statistics of the environment are flooded with lots of blue and water and sky and so on there's very little blue we actually touch you try and reach out and grab the sky its color disappears or same with water and then we talked about the high correlation between the fmry color tuning functions of voxels down it and object color probability which indicates or suggests that it is organized by this color but using color information towards animacy judgments and that color is sort of feeding into it in lots of different ways one of which is for that sort of animacy or or coarse scale object and maybe an intentional mechanism and then these color bias domains that are really privileging color over luminance contrast that seem to be using color to do something slightly different and those results then relate to the moving around the slide to the bottom right going kind of clockwise to the color statistics of objects and that the observation that that they are not uniform that the colors of objects are actually quite specific that objects tend to be warmer colored compared to backgrounds and animate objects tend to be distinguished based on the the prime or the yellow blue axis and those color statistics data interestingly enough are similar for for artificially colored objects which is I think instructive because it tells us that when you go about making the color of an object you borrow on this kind of implicit knowledge of what colors objects should be in order to paint those objects so we make stop signs red because natural objects tend to be warmer colored and we want that to be a salient object then we did these fmri guided microelectrode recording experiments to test the predictions made by the fmri and discovered that face selective cells do have this warm color bias have this ramp tuning function through the lm axis which suggests that these cells are specifically using the lm ratio to tell us something about the dynamic or changing character of of of faces chromatic color faces and then finally I talked about the paradoxical impact of memory on face color appearance under these conditions where retinal mechanisms for color are impaired people see faces as green and that and that that color match is selected really for the lm axis and I think tells us something important about the special role that color plays in social communication and with that I want to thank all of the people that have contributed to the work over the years I've listed them next to the specific projects I didn't talk about our magneto encephalography work or our work in in v4 if you're interested in reading up more on the right top right of the slide is the selection of papers that I drew upon to create the talk and of course I want to thank our funding agencies especially the national institutes of health the national eye institute intramural research program and you for your attention thank you so much thank you so much bevel that was absolutely fascinating and I'm sure you're getting lots of virtual applause across the globe um yeah that was really fascinating I really enjoyed hearing about um sort of language development and the behavioral relevance to color as well so there have been a few questions dropped into the chat which you may have covered as your talk went on but I'll bring them up anyway in case you wanted to add anything so the first question was from Wei Li asking are the three failure representing patches in it sequential or do they parallel parallelily receive inputs from v4 oh that's a good question yeah so the track tracing data indicate that there's lots of interconnection between the patches so we have we're right now in the process that's gearing up to do some functional connectivity data collection we've done a little bit of you know if you sort of look at the patterns of fmri activity in without driving it with visual stimuli you can look at which voxels are correlated with other voxels just in a kind of resting state and we've used that kind of an analysis to look at whether or not these bits of brain are connected to each other or at least driving correlated patterns of activity but um Wei Li's question is specific about whether or not that pattern of connectivity is hierarchical in the way that we that may may be assumed given um they're sort of our organization along the putative hierarchical visual processing streamed MIT and my suspicion is that the answer is they're not organized strictly hierarchically that there's lots of feedback connecting between the between them and there is evidence uh from track tracing data not guided by uh the functional data showing what the what the pattern of connections is and it does look like you get v4 input through a lot of it maybe not quite as far as interior it but certainly um through central it okay great thank you very much so the next one's from Tom Barden and Tom asks could the warm color bias be linked to the fact that it's the midget system providing the information which is numerically dominant especially in the phobia i.e the brain simply gets way more warm color information than information about blues or short wavelengths um that's potentially true that the problem with that kind of a theory is that we know in primary visual cortex at least develop kataras and devaloy have argued that there's an amplification of the escom system and that there's quite a lot of there's much more uh um there's much more responses to escom stimuli in primary visual cortex than you would expect simply given the relative ratio of escoms to lm cones so to the extent that the warm color bias we find is outside of primary visual cortex beyond primary visual cortex it i think reflects something active that the visual system is trying to do to recover that information um that said um super team re and colleagues have shown that there's very high gamma in primary visual cortex to warm colors to long wavelength sensitive long wavelength colors um which is consistent with the idea that you've got a lot of drive to you know red the problem with that hypothesis is that the midget system is not driving simply um the l axis of the lm uh uh sort of you know that in your cone opponent axis the midget system isn't responding selectively just to the l pole it's actually l versus m and they're not two independent midget systems and m minus l and an l minus m they're actually feeding in in parallel to the cortex so there's no reason to suspect that the midget system by itself would give you a warm bias it would give you the ability to distinguish warm versus cool and uh that i think is true i think that is in fact where these evident where these data lead us is down the idea that that what evolution has done in primates is created a midget system that gives us access to that specific access the lm axis um and the the piece about the color statistics of objects was really to underscore the point that that the um the color space reflects the extent to which we use color information and that is probably not an independent force from the force that drove the evolution of l and m cones um for obvious reasons the selective pressures are the things that made us get that cone system um great thank you um so we've got a few more so i'll keep going and then we'll we'll move into the zoom room um so the next one is from phil bartel and he says great talk have you plotted your data against wilkins and sario's vividness should we all become highly morphists and stick to the four causes um what was the end have we plotted it again to vividness and what was the last part of the question should we all become highly morphists and stick to the four causes i don't know what that means okay i'm i so i'm going to take interpret the question as something to do with saturation that's usually what people mean when they mean vividness saturation is a very so i haven't talked at all today about saturation and how saturation is encoded and in part it's because of the three standard dimensions of color that we appeal to you know hue that is the thing that you use for a color term red orange yellow green blue and so on luminance contrast the lighter blackness uh and saturation saturation is the least well defined um so it seems like saturation is the thing that maybe the brain is sort of most squishily adaptable to put another put another way i can take a given patch that looks very desaturated under one condition change your adaptation state and have it appear very very saturated um and this is nowhere more evident than in memory colors themselves where you know the the colors you imagine in your mind's eye are extremely vivid um and yet there's no saturation that you can measure at all um i think you know for those people who are interested in black and white photography i have a feeling that that's one of the compelling things behind black and white photography is that it allows the coloring of those photographs to be left up to your imagination which allows the colors to be really really saturated um in the development of color photography one of the things that drives photography technology most strongly is the desire for more and more and more saturated colors so we see this for hd televisions that now have you know more than three color primaries because they're giving you so much more saturation so there seems to be a kind of almost insatiable appetite for saturation but that itself confounds the ability to measure it great thank you um so the next question is from george and he asks does the separation of objects and backgrounds and animate and inanimate objects reflect primate visual space statistics or is it supposed to be a universal principle um so the distinction of uh that's so the object color statistics analysis we did um reflects primates in so far as primates humans specifically were the ones who labeled the images so um i i actually think you know if there's this quirky irony that we call these things objects they're the least objectives thing you can imagine they require a human observer to define what it is you know they're when i look out you know at the garden there's not you know a set of discrete objects that are isolated from each other that i can be like that's objectively defined it's like no i've decided to call that thing an object even though half of it is occluded and i don't know what the texture is on this side and so on um and that in fact was the point of using human observers to parse the images to label the images was to figure out you know if we're dealing with uh infrotemple cortex a part of the brain we think is involved in cognition that is in using visual information to connect to behavior then we need to have some some lever on behavior and um and so the answer to your question is it's specific to primates specific to humans in so far as humans were the ones that defined what an object was um and i suspect you could do the same analysis and people uh have done similar sorts of analyses where you look at you know what does a be considered to be an object and then you relate what it considers to be an object given its foraging behavior to the spectral sensitivity of its photoreceptors for example and it's a similar kind of spirit that's kind of what we're doing except instead of relating it back to the photoreceptors we're trying to relate it back to infrotemple cortex and the reason we're doing that is that we know that the photoreceptors in primates are not just for color vision they're supporting all vision you know cone photoreceptors are supporting black and white vision and motion perception and face perception and everything and so to ask the question about the relationship to statistics of statistics in behavior to brain activity we have to go to the right part of the brain where we think that relationship may be meaningful and that i think is in infrotemple cortex great thank you okay and the final question before we move over to the zoom room it's from jenny boston jenny asks for the color bias regions in it do you find any differences in color bias or other preferences between the posterior and anterior regions um we so the anterior regions what's interesting about the anterior regions is that they tend to show more overlap with face patches so the further anterior you get the less isolated are the color bias domains from the face patches they sort of seem to kind of have a little corner of them that overlaps and that's consistent with with the color tuning that we measure with fmri that shows that as you move for further anterior you might get more of a warm color bias in the in the in the color bias domains we don't see any difference in the face patches between that posterior and anterior so so what's happening in the more anterior domain seems to be maybe inheriting that color biased color bias for warm coloring warm color bias that we see in the face patches um i should say that um the resolution of the fmri especially in that temp anterior temporal lobe gets weaker and weaker because it's quite far down and it's a small chunk of tissue and so um our signal to noise also goes down uh and so you know there's a bunch more work to to be done and i would be curious to talk to jenny to figure out if she's got a specific hypothesis about what she expects to be happening down there but yeah good all right thank you so much that's all the questions um pleasure so far so i just want to say huge thank you um to bevel for that fantastic talk and also to the audience for attending and asking all those questions that was great um so as i mentioned in the chat bevel's available for a more informal chat so you can join us on the zoom link which has been dropped in the chat um and i also wanted to mention that the next lever hum seminar will be Joel Pearson on wednesday the 26th of may so i hope to see as many of you there as possible all right thank you so much great all right well i'll stick around for a little