Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Visual Perception and AI Inference | Penn Professor David Brainard
13 Plays3 months ago
How does our brain turn limited sensory data into a rich visual experience? In this episode, we sit down with Dr. David Brainard, a professor of psychology at the University of Pennsylvania, to dive deep into the world of visual perception, inference, and computational modeling.
Dr. Brainard shares his journey from physics to psychology and explains how his research explores how the brain processes light, interprets color, and makes inferences about the world around us. We also discuss his work in computational modeling, which is transforming our understanding of retinal function, neural efficiency, and AI-driven vision systems. From cutting-edge research to the future of brain-computer interfaces and neurodegenerative treatments, this episode is packed with fascinating insights into how we see and how AI is learning to see.
Recommended
Transcript
Introduction to the Healthcare Theory Podcast
00:00:00
Speaker
Welcome to the Healthcare Theory Podcast. I'm your host, Nikhil Reddy, and every week we interview the entrepreneurs and thought leaders behind the future of healthcare care to see what's gone wrong with our system and how we can fix it.
00:00:13
Speaker
Today we're speaking with Professor David Brannard at the University of Pennsylvania.
From Physics to Cognitive Neuroscience
00:00:17
Speaker
So hi, Professor Brannard. Thank you so much for coming on today. Yeah, it's great to be here. Thanks for reaching out. Of course. And I'm super excited to get into your work and research. But before we get into all of that, I mean, what's your background? What brought you into cognitive neuroscience and perception? What's I'd love to hear a story. Right. So I was a um back when I was, you know, your stage of my life, I was a physics major and I kind of interested in physics.
00:00:44
Speaker
ah And I would say that physics at that time, at least the parts I was interested in kind of conceptually were very much big science. And there's great things about big science, but I wanted to find a, I was excited about science and wanted to find a direction that was a little bit um smaller in scope ah where where you had a little more input to both the theory and the experimental aspects of things in a single lab, as opposed to kind of the specialization of science.
00:01:15
Speaker
particle physics and theoretical physics. And so I actually worked for a while after I was an undergraduate programming computers. um I traveled for a while and then I kind of ended up going to grad school in psychology just because of one course I had taken as undergraduate that had been exciting to me in cognitive science.
Finding Focus in Vision Research
00:01:32
Speaker
And I chatted with my professor and asked, you know, how do i go on in this? And he gave me some recommendations of programs and a little bit random, but I ended up...
00:01:42
Speaker
um finding a faculty mentor for my PhD, and he was studying vision. And um I liked him and his style of research. And that kind of got me into vision. It was less wasn't all that intentional that I ended up quite studying color, but that's what I've done.
00:02:00
Speaker
Yeah, and that's a great story that you kind of navigated your way like more organically. It wasn't always something you were um that you needed to kind of study or be passionate about. But i mean, what was, like out of curiosity, what was interesting about his research style and the work that he was doing that journey? Yeah, so i think that I think that some of it was you know in the study of vision and the study of perception um, we're able to make measurements that are actually quite precise in the realm of measurements and psychology. So that sort of resonated to what I had learned about in physics measurements. And in fact, uh, study of vision, it has its roots in people like Isaac Newton and, um, Herman von Helmholtz and, uh, giant clerk Maxwell, all of whom are known for their work in physics, but actually all of whom studied, uh, color vision as well.
00:02:45
Speaker
And, uh, So that and somehow resonated to the intellectual training I'd gotten as an undergraduate. And then I found um the questions of perception just captivated me.
Perceptual Inference and AI Connections
00:02:58
Speaker
And I think that kept me from, as you said, it was a kind of organic process, but it kept me from continuing my search for what I wanted to do. I, I, um,
00:03:09
Speaker
A kind of core aspect of questions in perception is how we make use of somewhat limited sensory information to make accurate inferences about what's around us in that inferential process, which is at the heart of ah perceptual psychology, but also at the heart of computer vision and in many ways of these the current trends in AI.
00:03:30
Speaker
how how that worked and how the brain managed to accomplish all it does for us. Just grab me as a set questions. and Awesome. Yeah, that makes a lot of sense. So is that, I think it's really interesting. Of course, like we don't need a full picture to like develop on it and ourselves. And it's a very like interesting process.
00:03:46
Speaker
So yeah, I mean, what are like the main questions you're answering? Is it more just about being able to really understand like inference and that area is there other areas that you've been able to explore so far? Right. So inference inference and how the brain makes inferences is a kind of big picture goal that I've pursued.
00:04:04
Speaker
um but But to get there, I have found myself, um I mean, in some sense, to make an inference ah from, if we think of the perceptual system as gathering sense data about the world, you need to understand what the data are that the brain begins with to give us our visual representations. And that then begins with light,
00:04:25
Speaker
and ah the formation of the image on the retina, so optics, and how the retina senses light, and how precisely those measurements, how how noisy are they? All those things kind of shape how we think about the inferences the brain has to make. And so I would say my work has had two threads, and one is on the side of thinking about models of um what we might call perceptual inference, and particularly about how we infer the colors of objects.
00:04:57
Speaker
But the other has had to do with, um and in the context of sort of your interest in healthcare, care maybe is the more relevant part.
Imaging Techniques and Retinal Research
00:05:03
Speaker
How does the eye and the retina work? um And these days I'm working hard to, um in collaboration on some methods for imaging the retina very precisely and studying the precision of the representations at a very fine scale behaviorally and how precisely we represent color information and then building ah actually a kind of computational model of that early initial visual representation is is occupying a lot of my time to really capture what we know about that, you might call it the brain's data acquisition process.
00:05:40
Speaker
Yeah, that's great. And I'd love to touch on like your comp like the work you're doing, computational modeling. But before we do that, mean, can you kind of walk us through like what is so interesting about light? Because there's a few key things like, of course, it impacts like at six out, like seasonal affective disorder. I know you've had some research there and then also being able to like understand how it's like influencing circadian rhythms. Like there's a lot of implications for how light can influence us in our daily lives. And but we probably don't know, like what are some surprising like discoveries that you've had or things that you've been able to research in that area?
00:06:10
Speaker
Right. So, you know, I think we're used to the idea, maybe this is where you're going, that we use light, it forms an image on the retina, that image is the first step of seeing the world. And now you're describing um a number of what we might call non-image forming functions of light. and And one of them is this, we have circadian rhythms, we wake up in the morning, or maybe not so early, however it goes.
00:06:34
Speaker
But we have almost everybody is on a sort of biological clock with an approximately 24-hour cycle. But that circadian rhythm gets entrained to the light coming up, getting light and it getting dark. And the cells and the processes that do that have really been the topic of a lot of discoveries, quite surprising some of them over the last 25 years.
00:06:56
Speaker
And in particular, discovery that there were some light-sensitive cells in the retina that we hadn't known about previously, that seem to be involved with that circadian entrainment. They're involved with the control of how large your pupil is, ah which is some work we did in humans. we We looked at how different wavelengths of light contribute to the pupil size and then not so much the circadian rhythm, which but but how maybe these newly discovered cells do or don't contribute to perception, which is an area that
00:07:28
Speaker
um we and a lot of people would really like to know the answer to. It's quite tricky to address in humans um because we can't, in in animal work, you can get a hold of animals who don't have the other kinds of photoreceptors and isolate these things. yeah People, we don't want to do that.
00:07:46
Speaker
So it's trickier. ah But we've been we've been trying to understand how these newly discovered cells, what what visual functions they contribute to. Yeah. And can you, um it would be like, that's that's super exciting. It'd be helpful if you could maybe, you provide like an example with specific area, whether it is like circadian rhythm and like that, um like those cells involved with that, like what is an example of like one thing that you're studying how'd you kind of go about that?
Non-Image Functions of Light
00:08:10
Speaker
So I mean, actually the one we had started with, with these new cells was to try to understand their role in the control of pupil size. And so we, ah To do that, we had to develop an apparatus, a little infrared camera that would measure the subject's pupil size. And then the trick we use to do this work is that the newly discovered cells, their their relative sensitivity to so different wavelengths of light is different from the classical photoreceptors.
00:08:38
Speaker
so that by really carefully tailoring the spectra of light that we shown to our subjects, we could emphasize the responses of different classes of cells in the retina and the exact details of how we you know choose those spectra are a little bit involved, but they basically capitalize on knowledge of different light-sensitive cells being differently sensitive to different wavelengths.
00:09:05
Speaker
And that allowed us to um show, for example, that these newly discovered cells, although they can contributed to the control of pupil ah They did it on a very slow, long time scale.
00:09:18
Speaker
And then the more classical photoreceptors could cause the pupil to dilate and contract um more rapidly. And so its we think that the these so-called intrinsically photoreceptive retinal ganglion cells, the newly discovered class, and lots of lines of evidence that support this, are averaging the overall light level over a longer time period, which makes a certain amount of sense if what you want to do is control a circadian rhythm.
00:09:45
Speaker
You don't want to be you know driven too much by a short flash. You want to average it over time, and that also makes a certain amount of sense for pupil control. um And so we we were able to kind of dissect a little bit, ah virtually dissect how different ah paths by which light information goes through the retina ah contributes to the pupil.
00:10:06
Speaker
And interestingly, there's there's actually one class of cells that makes the pupil, when you add more light, The pupil gets, it's counterintuitive, it gets bigger rather than smaller.
00:10:17
Speaker
yeah a like That was a very paradoxical little result. But it's generally swamped by the other cell types. It was only with this very careful dissection that we were able to see that.
00:10:29
Speaker
And that then matches up actually with some measurements other people had made of the physiological inputs to these cells in, again, an animal model. So... We were able to kind of link the animal work to the human performance, which is, ah you know, as we try to understand the neural mechanisms of seeing and think about things that go wrong and say, the retina and it would impact vision or other functions, making those links ah to the human ah is is a kind of central part of the thinking that we do.
00:11:01
Speaker
Yeah, it's interesting how you can draw that connection to like animal models. And I think like I had that observation a lot because photoreceptors, I guess retinal ganglion cells are able to have... Aren't influenced as much like by like volatility of light and like perception, I guess, which makes sense if you're tying it to circadian rhythms. but I mean, that's an interesting place that's, of course, like you guys are still having a lot of discoveries and it's rapidly developing.
00:11:25
Speaker
mean, just 10 years ago, I'm sure your understanding was pretty different than it is today. And a huge part of that comes from computational modeling. Can you kind of walk us through like what even is that? Like what are the applications and like, yeah.
Computational Models in Vision
00:11:36
Speaker
intellect so So computational modeling, I think, is a very broad term. Yes. yeah In our hands. um Well, so maybe I'll focus on what we're trying to do, which as an example,
00:11:48
Speaker
um We're very focused on, so so let me back up and say, light enters the eye through the pupil. And actually, even before it reaches the retina, it's blurred by the optics of your eyes. So, you know we're both wearing glasses. Those glasses are doing a little correction to features of our optics, which aren't quite um as well matched up to focus light as we might like them to be.
00:12:11
Speaker
and But everybody's eye blurs the image a little bit. um And so we in that blurring actually is one of the things that limits our resolution in the end. but you know The finest things we can see ah are are limited by how much the the retinal image itself is brewed. So we model that image formation process ah by just a set of little calculations that the computer can do. We start with an image, say, on a screen.
00:12:39
Speaker
We ask what's the image on the retina, and we build on a lot of work that people have done to characterize how blurry are the eyes, optics, etc. And then we know that that retinal image is sampled by an array of photoreceptors. We know a lot about how many they are and what wavelengths they respond to and how much light it takes to produce a response and how noisy the response is. And we spent a long time basically writing computer code that did a couple of things. One was it took a lot of data from the literature, which was in different places and sort of brought it into one computational package.
00:13:14
Speaker
that ah provided a model of a typical human retina where we kind of got all the sizes of everything right and the number of cells and matched it up to the optical blur. And that lets us calculate um how did the photoreceptors respond to light?
00:13:28
Speaker
um And that's a, so that's a computational model of just really, but you know, it's the very first little bit of the nervous system, just those photoreceptors, which took us a surprisingly long time to do. And then we use that in different ways. We asked, well,
00:13:44
Speaker
How much does the optics blur the image and how much does that limit performance? And we could simulate out how well could a, ah you know, ah we call it a computational observer, but a calculation that had access to our simulated responses.
00:13:59
Speaker
How well could that thing see? How well could it do specific visual tasks? How did that compare to the way people do them? And one of the nice things about the computational model is then you can say, well, suppose the optics were better.
00:14:11
Speaker
How much better would you expect to be able to do? Suppose they were worse. Suppose you were nearsighted. And we could look at the effects of those things. And I would say one of the goals of this model now, as we're moving in further to the retina, is to be able to understand um how different pieces of the retina contribute to visual performance. So we know there are different kinds of retinal cells that take the responses from the photoreceptors send them to different parts of the brain.
00:14:42
Speaker
As we work to model those, Our hope is that we will be able to understand how each one contributes to performance by basically turning them on and off from the model.
00:14:53
Speaker
ah yeah And when you're thinking about something like, well, maybe I should stop and see if you have questions about that rather than steamrolling on. No, actually should keep going. That's interesting to hear how your application is kind of turning out.
00:15:07
Speaker
Yeah, on that note, so like, I guess actually one question that I was thinking of, and I think that's important for the audience to hear is that, I mean, this is of course, like, as we're going to go into the future, it's going to change a lot, like how we view and use computational modeling. And of course, um but I guess the the best way to understand this ah is like, really asking you, like, how's that changed in the past, in the past, like, 10, 20 years, and even right, right?
00:15:28
Speaker
No modeling, like, what are the major advancements? And what does that really enable to you as team? Right. do So I think the biggest advance with faster computers in in our work is the ability to move from what we call, I would call it modeling very specific devices ah kind of experimental tasks that we might look at.
00:15:50
Speaker
And now having enough computer power and having put together enough pieces that we could play real images or real movies through the model and look at the outputs that that would better capture how we see when we look at stuff in the real world.
00:16:06
Speaker
and And where I was going with this is to say one of the goals of the work, again, to bring it back to health care, is you know we know that certain blinding diseases affect certain classes of cells and so we'd like to have a handle on how if there's damage to one class of cells what deficits and performance do we expect based on the computational modeling which we can then turn around and say are those deficits or would that be a looking for those deficits would that help diagnose the progression of disease or there are many really cool techniques being developed by
00:16:42
Speaker
you know, colleagues across the country and the world who are developing therapies that might halt the progression of retinal disease or reverse the progression of retinal disease. And then you can ask, well, what are the most effective places to target those therapies that we would expect to improve vision?
00:16:59
Speaker
and what Because we can simulate those kinds of things in the model. Or if I build a retinal prosthesis, which would be a device that would electrically signaled ah stimulate cells,
00:17:11
Speaker
ah you can't always arrange that to quite mimic the natural thing. What are the consequences of a slightly unnatural stimulation going to be? We published a paper that showed that, uh,
00:17:23
Speaker
you You didn't get as much improvement as you might hope if you stimulated cells in a way that didn't match what would happen to them in natural viewing because you didn't quite have the control. But we can investigate those kinds of things with this sort of model.
00:17:37
Speaker
So in our hands, I think it's the scale of the models that have really changed and also just the precision of the knowledge we have about the retina, which we can then take advantage of in building these models.
00:17:52
Speaker
Yeah, it's that's really interesting. And I think one thing I'm curious about is like really the implications, like as you move down, to I was able to work on um through my internship, like a rare disease, like ophtalology ophthalmology drug. And it's super interesting how it's, it's of course, very tied into the fundamentals. But with your work in like inference and your work in visual processing perception, like where do you see the biggest implications being, um especially with inference? I think it's a less tangible concept, maybe bit easier to understand. Like where do you see that kind of tying into? Yeah.
00:18:21
Speaker
So I think, you know our understanding of these inference processes, those those are probably neural processes that are happening not in the retina, mainly. I think I sort of think of the retina as the information gathering um piece of the visual system.
00:18:35
Speaker
And probably the inference parts are happening somewhere in the visual cortex or the ah parts of the cortex where vision meets memory and knowledge of the world, et cetera.
00:18:48
Speaker
I think that, you know, at at this stage um of our understanding, i'm I'm going to call that pretty basic fundamental science. It's not that I think um we we have in the next few years in reach and understanding um the inference processes in sight that will let us say, oh, it's those cells. And, you know, in this in this form of neurodegeneration, this is going to be the consequences. But in the long run,
00:19:17
Speaker
ah You know, we we walk around in the world thinking and seeing and doing, and it's these inferential processes ah that that are probably not just sensory that are enabling us to do that. And I think the kinds of understanding, and this might be five years or 10 years or 20, don't know how many years out, but it's the foundations that would then uh, as people think about, Oh, can I put electrodes in the brain that would help reverse neurodegenerative disease or substitute for it? You really need to understand those more complex processes. And, you know, the hope is that as we both understand the behaviors that we successfully do, and, and then link those to the animal model work and understand the neural processes that eventually it would, um,
00:20:06
Speaker
ah perhaps enable treatments of more cognitive cognitive conditions or so-called, if you have a you know a stroke to party a visual cortex, you can lose um certain kinds of vision.
00:20:19
Speaker
Perhaps we would ah understand enough to build prostheses for those kinds of things in the longer run. Of course, yeah. It's really interesting to see how like that can tackle neurodegenerative disorders. and Because we we spoke with on this podcast book with Dr. Chad Botin, who's people who've been working like BCI and brain-computer interfaces. And it's interesting. And I think that when you're thinking about like perception, you have the retina, which is like the data acquisition like you spoke on. And then the inference which like is a bit behind that, or I guess in front of that, but um up in the cortex. And for you, like where do you see like the implications? And you touched on like AI and computer vision.
00:20:56
Speaker
Yeah. Very exciting feels. Where do you see that kind of tying into your research and our ability to build those?
AI Models vs. Brain Function
00:21:02
Speaker
yeah So, so that's a, that's a complicated question. Um, so I'm trying to, so i' I'm going to make a few comments. Um, uh, the first is that, uh, you know as, as everybody who's paying attention to the news reads almost every day as some advance in, in AI and sort of machine learning and, uh, tools that, uh, you know, now are, are remarkably, you know, we can kind of almost talk them, you can talk to them and they'll give us answers and, uh,
00:21:29
Speaker
do these kinds of reasonings. the The basic architecture of those machines um really came from, they're called artificial neural nets, and they came from ah an understanding of ah you know, or it's an abstraction of how the brain is wired up and does things better.
00:21:46
Speaker
The details of how they work are not quite brain-like. So i think the first thing to say is there was a real in kind of information flow from neuroscience into computer science and machine learning that was quite profitable.
00:21:59
Speaker
Now we have these AI-trained models that do various tasks. Some of them are visual models. And we are we are asking, are we being many people in the field, including to some extent myself, asking, are those computational machines good models or not so good models of how the brain actually stuff?
00:22:24
Speaker
And i think... The answer is, well, they're better models than we had 10 years ago. So that's exciting. they They capture certain aspects of performance and certain aspects of neural responses quite well. And then there are interesting differences in how they work and how they perform. And people are kind of digging into those. So you can, there's a line of research where you can demonstrate, you kind of but take a trained AI model.
00:22:52
Speaker
You can, um, uh, develop what they call adversarial attacks on it that cause it to give very bad answers by very small perturbations of its input.
00:23:03
Speaker
And it's harder to do with the brain. And so there's something robust about our the way our brains work. Our brains use a whole lot less power than these artificial neural networks. you know, we sort of are running on about a 50 watt light bulb or something terms of the brain's power and not not the entire output of a hydroelectric dam.
00:23:25
Speaker
ah and the amount of data that we don't, it's a little hard to say how much data have trained us. But um anyway, so there are, i think now a real interest in looking more at why is the brain so energy efficient and in its computations and why is it robust?
00:23:40
Speaker
in ways these network models are not. And I think there's going to be a real healthy and continues to be an interplay between the engineering development of ah machine learning things then that just perform in ways that were really unimaginable 10 years ago in terms of what they can do and the extent to which they are good descriptions of the brain's computations and not and how the deviations will maybe tell us both about the brain and about how to put better um AI.
00:24:11
Speaker
Yeah, it's it's very interesting to see like how at this point with AI, like with not only open-anthropic, but like the biggest like research labs of big tech companies, are like the idea of industry and academia and research are kind of merging together because AI is such an evolving field. And um i mean, you've been in academia for i mean most basically most of your career. And I think that this podcast is really more oriented around entrepreneurship, seeing how we can take our insights to the real world. But um like What has really been an academic and getting that foundational approach really brought you in your career and like what has been rewarding about it for you?
Curiosity-Driven Research in Academia
00:24:49
Speaker
So, you know, I i have to say, you know I was pretty early in my career. i was mainly driven, not entirely, but I was largely driven by questions that I just personally found fascinating and pursuing the answers to those questions. How do we see the color of objects?
00:25:04
Speaker
I want to know the answer to that question, which involves. ah ah Kind of a complicated inference because the color of um the information we get about object color turns out to be affected by things like what's the illumination, what's the color of the illumination and how does the brain separating these things and it wasn't.
00:25:22
Speaker
i think my interest really was just driven by curiosity and I would say that. um
00:25:28
Speaker
curiosity-driven fundamental research is something that is easier to do in an academic setting um where you are not having your deliverables quite so much in terms of how can we make this product better or what disease is it going to cure.
00:25:46
Speaker
And um that that, I think, is you know the role of academic research in many ways is is to be thinking about questions on a time horizon that's longer than a product cycle or longer than curing a disease in the next few years.
00:26:01
Speaker
And it it provides for discoveries. And it's often hard to, it's hard to justify because you don't, you can't quite say what is it I'm going to discover and why is it going to be important later on?
00:26:14
Speaker
um But it's easy sometimes, but you can look back and say, oh, you know, the reason we can put people in brain scanners and non-invasively diagnose tumors is because some people were interested in how, um you know, molecules absorb magnetic energy and and how that worked. And they were just interested in it as a fundamental physics thing. It turned out to have enormous implications for medical technologies. 20, 30 years later,
00:26:42
Speaker
um And it's sort of hard. So it's hard to know quite what basic research is being done now is going to discover. One of the things I have found is that in pursuing those academic interests those questions that fascinated me, I had to develop techniques for making ne measurements of behavior. how How did people see color? or How precisely did they see it?
00:27:07
Speaker
How did the context in which they looked at something affect it? And ah so i I, in some ways, became over my career reasonably expert in quantifying perceptual behavior, which was something I had to do to get at the basic science questions I was interested in. and And what I have discovered is that I have later in my career had a number of collaborations where my knowledge about how to do that turned out to have great applicability to much more applied research.
Collaborations and Clinical Applications
00:27:40
Speaker
problems. And so I collaborate with a neurologist, Dr. Jeff Aguirre quite a bit, who did the work on the um intrinsically photosensitive system with me.
00:27:53
Speaker
And he was interested in that from a neurological point of view. And some of the techniques that I knew about from studying color just turned out to be the right ones to get at those questions. I collaborate with um Dr. Jessica Morgan in ophthalmology, who studies um the retina and imaging the retina using very advanced, super cool techniques and actually techniques that now allow us to do behavioral experiments at a really high retinal resolution.
00:28:21
Speaker
And again, it's sort of my expertise in the behavioral side of it, which is now allowing us to understand the functional implications of certain kinds of structural deformities that occur during disease and really get at more, ah we'll call it translational work. ah Work I did on understanding color inference,
00:28:40
Speaker
and And color cameras led to a project on a clinical trial for a drug that was intended to to cure an eye condition, an eyelid condition that causes the eyelids to be inflamed and diagnosed by how red they are.
00:28:54
Speaker
ah turned out things I knew about color were just useful in evaluating the efficacy of this treatment. So it's a kind of like a... spin-off of the basic science that the measurement techniques turn out to have all kinds of applications um that surprised me in an extremely pleasing way in my career. it wasn't what I had set out to do, but it's rewarding to me that the some of the work I've done has had very tangible, I'd say, contributions to health and in some cases, technology, how cameras process images and things. In fact, the inference work
00:29:30
Speaker
has had more applications in the engineering side than precision of sensory measurement work more on the health side. Yeah, I think it's a great place to be where you're driven by that curiosity, but you're still having tangible impacts outside of like um what the immediate like research that you're doing, which is great. And I think that thing that's super exciting is being able to quantify things that are not easily quantified, like your work in vision. like For example, a little off topic, um was like working in psychiatry and like
00:30:02
Speaker
neurodegenerative drugs. And it's really hard to quantify ed endpoints and like clinical endpoints and clinical trials because it's it's like there's no quantification to it. So I think it's a very interesting problem to solve that can have um yeah huge implications. And I guess a couple of quick like fun more questions, like now that you're kind of moving forward in your career, I've accomplished a lot. like What's the maybe biggest unanswered question that you're still looking to solve in the future?
00:30:27
Speaker
You know, I i happen to know the answer to that question. Yeah. Well, we like to we'd like to push this model. our Our goal with the computational model I described is to um have a reasonable model of every bit of neural information that makes it to to visual cortex from the rep map.
00:30:45
Speaker
And I don't know that we're going to get there know what's left of my career, but but that's a real goal for me. There'd be a real, I think, a real milestone. And, you know, our model isn't perfect. But it's a foundation and it's something other people we hope will pick up on. So that's one thing.
00:31:00
Speaker
ah The other thing I'm working very hard on these days, and actually it it benefits from some AI in the sense that AI techniques are helping us make the guide, the measurements more efficiently is to try to characterize how precisely we see colors for every possible color.
00:31:18
Speaker
ah and every possible change from every possible color. So you know there's a question people sometimes might say, how many colors can we see and how many different colors can we perceive, that but which is in the end answered by how different do you have to make the stimulus before you can tell it changed?
00:31:37
Speaker
We want to make a better set of measurements ah directed at that question that are more complete ah which we think we can do as a sort of foundation for really understanding the precision of color vision and understanding how we tell, um how we judge how different any pair of colors are.
00:31:57
Speaker
So we're we're very excited about that. We're chugging away for it. I just submitted a grant proposal to ideally get some funding to ah take that project to completion.
00:32:10
Speaker
Of course, that's really exciting. And I think, I mean, best of luck with the grant thing. That's going to be exciting. It's also cool to see how it's not just coming up with an answer that's important in your world. It's about coming up with the questions that you're trying to answer. That's just as important. But I've really appreciated the time get today. It's been super helpful to not only walk through your story, but just see where you're working on and see how that research can have like such interesting implications and have such a wide um like tangible effect. So really appreciate your time and thank you so much for coming on Health Care Theory. My pleasure. And thanks for um your work got just ah communicating about science and health care to everyone. I think it's really important that we try to explain what we do and why we do it. And so I really appreciate your time as well.
00:32:58
Speaker
For it. Thanks for listening to The Healthcare Theory. Every Tuesday, expect a new episode on the platform of your choice. You can find us on Spotify, Apple Music, YouTube, any streaming platform you can imagine.
00:33:11
Speaker
We'll also be posting more short-form educational content on Instagram and TikTok. And if you really want to learn more about what's gone wrong with healthcare care and how you can help, check out our blog at thehealthcaretheory.org. Repeat, thehealthcaretheory.org. Again, i appreciate you tuning in and I hope to see you again soon.