10:05:31 So we can only go up here from. Remember, the rules of the game is that you have 90 s and one slide that you have already prepared and uploaded and we'll all be using this same computer.
10:05:47 All right, so.
10:05:51 It's my pleasure to be here. I am a theoretician.
10:05:55 I work on mostly on normative approaches to brain-minded behavior, which essentially means that I'm using a number of different methods, including Bayesian machine learning, reinforcement learning, deep learning all those methods that you can see there to try to address different levels of
10:06:16 organization of the brain spanning a large fraction of the continuum that goes from very low level kind of biological details to kind of high level, cognitive and behavioral.
10:06:30 Issues.
10:06:34 Some of the work that we do is at the level of individual neurons, and we are trying to understand how individual neurons integrate inputs both in time and in space and the general theme is that we are interested here, we have some other work that is more descriptive.
10:06:52 But here it's a normative approach and we are trying to understand these integrative prophecies of individual neurons in terms of how they are adapted to the statistics.
10:07:00 Know the buzzwords here of their incoming presynaptic inputs in the way they integrate those inputs in turn.
10:07:10 We also do some work at the level of neural circuits.
10:07:12 In fact, I guess these days probably most of our work is concentrated at that level, and there is different themes within that.
10:07:21 And that was my time up. So we work at the level of neuro circuits and at the level of behavior.
10:07:27 You'll hear more about that, and perhaps most importantly, I have some social interests that you can read.
10:07:33 There. If all goes well, my family will be joining me later.
10:07:36 That's still a little bit up in the air, and I'm very much looking forward to speaking to all of you here. Thanks.
10:07:42 Next one, Libya.
10:07:52 It's still sunny. Good morning, I'm Olivia Bell.
10:07:57 I work in Berlin. Charity is a fantastic neuroscience.
10:07:59 Community. Come and visit us, I'm very interested in what I call the emergence of statistical learning from suportico to cortical, and my interest is an experience dependent plasticity, and how it happens at different levels of processing what triggers.
10:08:14 It. So we study 2 cortical cortical loops, have put their diagram of rodents, brain.
10:08:19 We do this in mice and I'm particularly interested in the auditory system, which I think is optimized to do.
10:08:25 Statistical learning we're interested in loops between infiric colliculus and cortex through thealamus.
10:08:31 And we. My interest is in experience, dependent plasticity.
10:08:36 I, as I said, how it there's represented. What is an engram?
10:08:40 What is a memory generalization? We use it as a very interesting way to understand this representation, and cortical fuel feedback.
10:08:49 What is the role of this top-down? We do this add through different questions, contextual learning.
10:08:58 Both the roll of auditory and multimodal background in cortical, cortical, and hippocampal structures.
10:09:04 We record simultaneously from auditory cortex in fearicolicolis and thalamus, as well as hippocampus.
10:09:10 With near pixels probes, and we're looking, for example, at statistical models such as intensity, sound intensity, or sequences.
10:09:22 Or this predictability in the system, and we also have behavior, behavioural.
10:09:31 Behavioral task automatized behavioral tasks.
10:09:34 And when I'm feeling sociable I like to play ping-pong, dance, dangle, or go sailing.
10:09:39 Thank you. It's nice.
10:09:45 Ahmed. I didn't see him actually.
10:09:49 Joseph Chance.
10:10:08 Alright. So yes, I am the one who is doing statistical learning, but I guess is a the lines of the Hey rob paradigms which?
10:10:28 How you pick up any kind of structure of that. That's what you say for physical learning.
10:10:34 Do that to adults, honeybees, normo sets, and the main point it seems to be that people might do things equally well for people.
10:10:42 I mean this is, but they do it in very different, according to very different kind of other theme, is that.
10:10:52 They are kind of there system, or the second point is the physiology could be kind of pushing with some people in the room, and some people not in the room.
10:11:05 The idea that the brain does probabilistic computation, approximative proximity, computation, evidence for that working on the next and next step.
10:11:17 So far this is truly the case. Then can we find some more evidence of that?
10:11:22 The third thing is decision-making, which is a little bit more complex and usual decision-making.
10:11:33 Those 5 which is going to do. I'm going to do a different kind of learning, a different kind of decision-making, and I'm not going to tell you that only if you come to me person at the most important thing social.
10:11:49 While, I suppose, multiple location serving, saleing, swimming all this same interest me.
10:12:01 Last one know who he is next. Yeah, I know that right?
10:12:05 So!
10:12:13 And I am Jonathan.
10:12:23 T start. Okay. So John Victor, my deep background is in mathematics.
10:12:31 I'm at Wow Cornell, where I'm in the department of neurology, neuroscience.
10:12:35 And I got there via Phd. In visual Physiology.
10:12:39 Most of mice. Scientific life is been envisioned, but more recently some olf-fashioned.
10:12:48 II don't like lists. I do like tables and hierarchies, so I try to organize things into a table, not a list.
10:12:54 So one line of interest is the statistics of natural inputs.
10:13:00 So that means that natural scenes unnatural scenes, odors interested in neural computations, mostly mid-level vision, so visual texture, segmentation, objects, how things are organized into perceptual spaces and that's intentionally shaded down into the olfactory
10:13:20 domain, and more recently, active sensation. And this is with collaboration with Mckelli Ritchie at University of Rochester, and Fixational eye movements and high acuity vision is a particular focus, olfactory Navigation.
10:13:34 Antenna will move sniffing, and that sort of thing is another focus, and to complete the table, well, okay, so pursuing these things with some theory, some experiment, psychophysics and computation and a couple of things on the recreation, side, I have a few seconds left so just okay, so
10:13:55 recreation hiking. I want to try to organize something for Saturday.
10:14:00 Hiking is a multi dimensional area. So please let me know what level of altitude change and time change.
10:14:06 You're interested in, and we'll try to figure.
10:14:20 Alright great. My current focus is large-scale and multi-level models of brain and behavior by multilateral looking at that figure on the top left.
10:14:30 That's barorrowed from Carl Craver, philosopher of science and ideas at the top level.
10:14:35 You could think of it to its behavior. Usually in biology, the First level of mechanism refers to a biological, substrate, instead, I use a cognitive model or deep learning model that has that could capture behavior, but also has components to be related to brain function but importantly what these multi-level
10:14:49 approaches, you could decompose it to a lower and lower level.
10:14:52 As a scientific question demands. So like one example shown here on the upper right, we took this cognitive model of learning that had these high-level cognitive constructs, these clusters, and there's very few of those with very many neurons.
10:15:04 But we decomposed them, so we got rid of the clusters I just had thousands of neuron-like units that coordinated together through this neural flocking, relying on recurrence in the hippocampus to kind of make this multi-level account
10:15:16 from neuron like units. So like the cognitive level to something like behavior.
10:15:21 Large- we're interested in using deep learning models and also developing methods to better evaluate them.
10:15:27 So rather than you know, using correlation approaches like Rsa.
10:15:30 We use this direct interface, we say, if a brain region corresponds to a model layer, we should be able to replace model activity with brain activity and drive the computation forward.
10:15:40 And when you do that you get a different view of function also at the large scale, wrap it up.
10:15:48 Yeah, large scale looking at interactions between regions that's controller peripheral architecture to model coordination between regions like how higher level things going on in prefrontal cortex visual regions all right?
10:16:01 And I like hiking, too. Thanks.
10:16:13 Let me know when you're ready.
10:16:14 Go right. Mine is Ulrich by a home. I am from beautiful northeast England.
10:16:20 Durham University. I do computational modeling as well as some experimental work to mostly split into 3 kind of areas all kind of with a normative kind of starting point.
10:16:30 So we've done quite a bit of work on patient inference.
10:16:33 If you have different kind of multiple cues, visual auditory, what should an observer actually do in terms of combining these together?
10:16:41 Especially if they don't know things about the structure. For example, behind the underlying stimuli I've been doing work on reinforcement learning, especially in terms of things like bigger.
10:16:52 How much work should an animal likely put into trying to reach some kind of reward, but also in in terms of choice.
10:16:58 What kind of choices did you make which nicely leads into also work on new economics and decision-making?
10:17:05 So if you're trying to maximize, say an expected utility, what's the best way to do that?
10:17:10 If you have certain uncertainties about rewards and potential outcomes of your actions, which, of course, involves both elements of reinforcement, learning and beige, inference.
10:17:22 Experimentally, we've been using a large range of difficult tools, but usually focused around humans, mostly at adults.
10:17:32 Some work on children to using behavior. Ed, for eye tracking, eye tracking.
10:17:36 And even a little bit of pharmacology, and let me use the last 8 s to tell you that socially I do like any kind of games or games.
10:17:46 Bull playing games, video games, hiking music, and anything that involves being on in or off on the water.
10:18:07 It's start. Hi, everybody! So my name is Simon Rompey or Simon. You can call me.
10:18:13 I'm a trained neurophysiologist.
10:18:15 So I came from the Ltp. Field. Naturally I'm interested in plasticity in the brain, and how the plasticity can help us to learn and memorize things.
10:18:24 And so in my lap we had been looking at synapses, for example, doing a lot of imbibo imaging, using the mouse auditorory cortex as a model where, as an Ltp.
10:18:34 Person, my prior clearly was, brain is still. You set the tetanus.
10:18:38 Then things change. That's not what you see. When you look in the brain, lots of stuff's happening, even in stable environments.
10:18:45 Then basically that much of the statistics you see going on can explain steady states in the connectivity.
10:18:54 Looking at neuronal examples, the plasticity you see happening ongoing actually transmits to the activity scale.
10:19:01 You see, neurons change their tuning over time. Nevertheless, if you look at a large enough population of neurons, you see that pairwise similarities of activity patterns are preserved.
10:19:12 So a stable representational map is being built, and this map got me really excited.
10:19:17 And last year's because it's a gloriously failed in predicting what my mice can easily do, and that I cannot easily do.
10:19:24 But the the map of stimuli seems to tell us really how.
10:19:30 For example, generalization is happening. So you can predict generalization to complex sounds, even based on similarity of the map.
10:19:39 Apart from that also, I brought my ping pong paddle.
10:19:44 So I like that a lot. I like drinking wine and all the other things.
10:19:56 Awesome.
10:19:58 Athina.
10:20:05 I go? Hi, everyone. So yeah, I'm Athina Academy from the Established C in London.
10:20:11 Maybe I am a systems neuroscientist. So in the lab, we work with forest species which are human rad smice and computational models.
10:20:21 And we try to basically kind of look into interaction between our working memory system and I use the working memory in order to just refer to some internal representation of our sensory world.
10:20:34 And this internal representation should have some properties that would allow abstraction of rules, abstract learning of patterns and integration with our basically learned priors.
10:20:45 So we try to kind of study working memory in interaction with statistical learning when it comes to statistical learning, we do have some reinforcement learning tests that learning statistics would be beneficial Quintin is working on that he's going to present nineteenth second, slide then we do have
10:21:06 some other tasks, that there is no reward involved.
10:21:10 So it's not reinforcing. Learning, then, is going to present something on that. Briefly.
10:21:14 And then we also have some projects related to transfer learning how once we learn some knowledge, we can basically use that knowledge in other contexts.
10:21:23 So we combine behavior in that human and mice with some neurobiology in rod, its brain, and always kind of tied to some modeling works, either normative modeling works, either normative models or neural network models.
10:21:38 Thank you.
10:21:45 20.
10:21:52 So, Hi! I'm quintin. So I'm in Athena's lab, and I do mostly the mouse stuff.
10:22:00 Although they're trying to do some of the modeling stuff as well, even though that's more new to me.
10:22:05 And basically most of the stuff I do is taking perceptual decision-making tasks.
10:22:11 In particular, this auditory, decision-making task in mice, exposing the mice to alterations in the sensory statistics.
10:22:18 Looking at the effects of those alterations on the behavior, and then digging into what it is exactly that those alterations are doing to the behavior what the mice are learning.
10:22:28 Does it actually have anything to do with the alterations of the statistics?
10:22:31 Are they just learning more simple heuristics, and then implementing methods such as optogenetics, both fiber based and using a more sophisticated Java scanning system to look at the different areas of the of the dorsal surface and to see what areas are you know
10:22:48 causally involved in these behaviors, and then, more recently, we've been trying to set up new techniques to pair these manipulations with neural recordings.
10:23:00 And yeah, I, when I'm not working, I like running, although I will not be taking those running shoes to the beach now, because I'm unbelievably afraid of getting them stained with all that tar.
10:23:12 But otherwise, yeah, that's me.
10:23:28 Okay. Hi, everyone. I'm Danny. I'm also a Phd student in Athena's lab.
10:23:33 My project is sort of a crossover between, I guess.
10:23:36 Sequence learning and statistical learning. Where we're trying to understand.
10:23:42 Say, if I present you to sequence. Ymc. But then I present you a random fourth image.
10:23:46 Why use? I would say surprise. So the question we're looking at is based on your sensory history, right from events in your life.
10:23:55 How are you able to acquire this generalized rule? Of what a sequence, or what pattern should be?
10:24:02 So I have. I have this idea that structured events lead to a learned regularity.
10:24:08 And the thing we're probing is sort of identifying a proxy signal of this learned sequence.
10:24:14 And, by the way, I'm doing that right now is looking at using mouseuperilometry, trying to identify this surprise signal in the people diameter.
10:24:21 So we have some data on that. And moving towards manipulations.
10:24:28 And other than that, I am a big foodie. I run a lot and I swim. So anything active.
10:24:33 I'll be doing. Yeah.
10:24:49 I have animation, so I have to know how to do it.
10:24:51 Okay, good. Okay. So I'm Tim Brady. I'm coming from San Diego.
10:24:58 I brought broadly study, visual memory, and there's visual learning, and I think a lot about how learning and knowledge impact visual working memory and visual long-term memory.
10:25:07 And we also think a lot about sort of probabilistic representations, perceptual awareness, visual ensemble representations, and broadly, like a lot of people who study memory will be somewhere on the axis of like all the way studying colored dots are studying the real world, and we we tend to live
10:25:22 right in the middle. So studying things like, you know, arrays of objects that don't have full scenes in physiology and making models of how people represent sort of medium complexity things like textures, and how people learn regularities about you know which colors co-occur and a behavior we do a whole variety of
10:25:40 stuff that's with real scenes, with perceptual experts like radiologists with a how learning impacts working memory representations.
10:25:51 If you know the backpacks usually read, does that help scaffold your ability to know what luminance it was?
10:25:53 If the luminance is always different, something we've been doing recently, and I'd say that idea where we're most excited about these days is that we spend a lot of time thinking about how the sort of representational similarity structure representational geometry of stimulus spaces can help
10:26:06 predict performance in lots and lots of tasks of if you ask people to do like a continuous report, what color was that backpack you'll find that this this you know how similar the representational geometry of color plus a simple decision model predicts an awful lot of tasks and that
10:26:24 people can report that whole continuous probability, distribution of what color it was that seems to be predictable by that geometry, and I was maybe 5 s long.
10:26:33 Okay, that's it. Thanks.
10:26:39 Alright!
10:26:44 See, does this pointer work by any chance?
10:26:49 No. Okay, so my name's Aaron. Saez.
10:26:54 I'm now at Northeastern University. I was at Uc.
10:26:56 Riveride just days ago. And so my work really looks at broad aspects of learning.
10:27:04 So I I've kind of moved, whenever you look at the left there's Donald Kim who's basically sipping water through her tube, or his staring at a screen.
10:27:11 And basically, we're doing kind of classical conditioning to see what this should change, how he perceives stimuli on the right.
10:27:17 Basically I started doing more ecological stuff. II do training that will improve people's ability to hit fast balls.
10:27:24 And that you know, really kind of my research has gone from starting off as a computational neuroscientist to working on.
10:27:32 You know how to come up with more conceptual models of you know how you know reinforcement learning might be able to explain how you learn both what you're trying to learn in a task, and things that are paired with it, or really the reality that whenever we're learning anything our entire brain is
10:27:48 changing in complex ways that you know every participant in our task will be learning different things how do we understand this more broadly?
10:27:56 And now I actually create a lot of software. So I run the brain game center for mental fitness and well-being where pretty much all those topics I'm doing something in.
10:28:05 I'm really with. How do we better measure? You know these different dimensions of cognition?
10:28:10 And how do we train it? I like outdoor activities.
10:28:16 So hiking, kayaking, cycling, mine tasting is outdoors.
10:28:19 It can be so, and that's it.
10:28:27 Yeah.
10:28:31 Hello! I'm Elava Fisher, and I'm from the Hebrew University.
10:28:37 I'm interested, I guess, in. I'm a cognitive neuroscientist.
10:28:40 I'm interested in large-scale theories of perceptual learning.
10:28:45 For years I have been studying what's on the left part, which is the relative contribution of top-down task related context to our learning, focusing more on top-down.
10:28:59 Then I'm bottom up and coming up with their reverse hierarchy theory for claiming that our first perception is the outcome of the is the top down.
10:29:10 Is the high level representation, and in the last few years I've been interested in special populations.
10:29:19 People with their developmental difficulty, a typicality whom I think they're typical stems from their unique style of learning, largely of statistical learning and specifically, I focused on 2 populations.
10:29:37 People with dyslexia and people with autism.
10:29:39 In the last few years. Basically, I think that they have opposing patterns of atypicalities in learning people with dyslexia.
10:29:48 This is kind of think about it. Forget it 10 s. People with dyslexia forget faster, and people with autism update updates lower.
10:30:03 And you're welcome to ask more about it. Anybody plays Dennis.
10:30:09 Yay. Okay.
10:30:22 Hey? I'm Lauren. I'm at the University of British Columbia.
10:30:26 I decided to take a different approach with my slide, and not present a whole bunch of different stuff that we do, but actually to both pitch our main method.
10:30:33 We study young babies, we do behavioral work, but we also do imaging.
10:30:36 I've done people lometry with babies. If you want to chat about that, I'd love to, but I also want to pitch.
10:30:43 We have this new near system, which is very cool, high density in years.
10:30:46 We're recording pretty much the entire surface of the cortex with babies.
10:30:48 About 700 to 1,000 measurements. So anyone's interested in years with babies.
10:30:53 I'd love to chat about it. But I also want to talk about my favorite graph because I really care about learning in relation to processing, perceptual processing vision and audition as well as cross-modal processing.
10:31:05 So what you see here is that when babies hear a sound and they see a visual event, you have activity in the temporal and the exhibit lobes, which is what you would expect if they've learned that the sound predicts the visual event you see a lot of occipital
10:31:17 lobe activity, even though the visual event was not presented.
10:31:19 And this is what a control would look like. So these are babies who hear the sound, but have never learned that they're paired.
10:31:25 This is a very quick learning task, as you can imagine with human babies.
10:31:28 It's a few minutes in the lab. So just a couple minutes of passive learning that a sound predicts a visual event changes the excepital lobe.
10:31:36 Mostly loc is where we're recording from here to here, which blew my mind.
10:31:40 These, babies are 6 months old. They're learning rapidly, but they're also changing their perceptual systems very rapidly.
10:31:45 So this was a graph that launched a thousand studies for us, and really looking at the integration of perception and learning.
10:31:52 Early in social. I do anything in the water. I love to swim boats.
10:31:57 Anyone interested in trying, surfing. Let me know if I've only done it once, but I want to check it out again.
10:32:03 Hiking trail running, but also any type of running I'm a big foodie.
10:32:06 I love cooking, and I love film so.
10:32:20 So Hello, guys, I'm deja nemat. And I came with, Tell Dorabic who is my work and life partner, and we came from a Lyon, France, and we are cognitive neuroscientist and community psychologist.
10:32:37 We investigate statistical learning from a lifespan perspective, we showed that children are better than adults in some aspects of statistical learning.
10:32:47 We have a lot of studies on a typical development such as Autism and Adhd and Tourette's Syndrome.
10:32:56 We showed, for example, that threat syndrome kids are better in some aspects of statistical learning.
10:33:02 We also investigate the interplay between cognitive control or executive functions or prefrontal functions and statistical learning.
10:33:08 We showed that actually, there's a competition between computed frontran and statistical learning so weaker prefrontal functions and statistical learning, so weaker prefrontal functions and working memory can lead to better statistical learning so it's a very interesting idea and we think that
10:33:23 no sense to investigate learning without consolidation. So we have a lot of study on on consolidation we think that sleep is overrated.
10:33:33 We show that sleep is very good. We show that sleep is very good for memory, consolation, but not special.
10:33:43 So there's no evidence for sleep-dependent consolation.
10:33:46 That's a provocative idea. That's a much more interesting investigating waterpass consolation and locally dependent conservation in the Viking brain.
10:33:55 Thank you. And we add our doing that and football. So that's it.
10:34:05 So!