cover xiao jing wang

Xiao-Jing Wang on working memory and prefrontal cortex

  • cover play_arrow

    PLAY EPISODE


cover xiao jing wang
Season 2010
Season 2010
Description arrow_drop_down

Description

Why is the ability to hold something in mind, even briefly, the gateway to flexible cognition? Xiao-Jing Wang explains how attractor dynamics and slow synaptic reverberation in prefrontal cortex give rise to both working memory and decision-making. Subscribe for more from the Convergent Science Network podcast series. Xiao-Jing Wang begins with a deceptively simple argument: without the capacity to maintain information in the absence of direct sensory input, an organism is enslaved to its environment, reduced to reflexive responses. Working memory , sustained neural activity that bridges the gap between stimulus and action , is therefore the foundation of cognitive flexibility. Drawing on decades of lesion studies, single-neuron recordings, and computational modeling, Wang makes the case that prefrontal cortex is uniquely equipped for this role, thanks to its dense recurrent excitatory connections and distinctive neuromodulatory environment. The interview dives deep into the mechanics of attractor networks, which Wang uses as the theoretical framework for understanding prefrontal dynamics. He is careful to demystify the concept: attractor states are simply relatively stable states, not rigid black holes. What makes prefrontal cortex special is not persistence per se, even oculomotor circuits show persistent activity, but the capacity to maintain multiple stable states simultaneously and switch between them with brief inputs. This multiplicity is what a working memory system requires, and it emerges naturally from the nonlinear dynamics of strongly recurrent circuits. A key surprise from Wang’s modeling work is that the reverberation sustaining working memory must be slow, mediated primarily by NMDA receptors rather than fast AMPA transmission. This was not a design choice but a computational necessity: fast positive feedback makes the network explosively unstable, while slow reverberation provides both stable memory states and the gradual ramping activity observed during decision-making. The same circuit architecture that holds items in working memory also integrates evidence over time, producing the reaction-time signatures seen in prefrontal recordings during perceptual decision tasks. Wang also addresses the frontier challenges: extending local circuit models to large-scale brain systems, understanding how mixed selectivity in prefrontal neurons supports combinatorial coding of sensory, rule, and motor information, and reconciling the role of neural oscillations and correlations with the stochastic firing of individual neurons. His vision is one of building blocks , elementary computational mechanisms that can be composed into increasingly realistic models of cognition.

Tagged as:

About the author call_made

CSN Podcasts

Both the triumphs of humanity and its most evil deeds have resulted from collaboration. In a time where humanity is required to aspire to the former and minimize the latter, the question arises of how collaboration arises and why it fails. Surprisingly, this phenomenon, so central to who we are, is not well understood. Hence, a collaborative effort is required to understand collaboration in its full biological, psychological, sociological, cultural, and economic complexity and to translate this understanding into operational impact. This series of podcasts is one step toward achieving these complementary goals. The Collaboration Podcast presents interviews with people who are central orchestrators of collaboration in various domains including business, government, science, art, health, sustainability, and the military. The discussions were conducted by Prof. Dr. Paul F.M.J. Verschure and members of the Program Advisory Committee of the Ernst Strungmann Forum on Collaboration (https://www.esforum.de/forums/ESF32_Collaboration.html) during 2021 and had the goal to sketch a map of opportunities, challenges, and obstacles in human collaboration. The forum took place in May 2022, and now we would like to share this series of interviews with a broader audience. The full report of the Forum will be published in 2023 by MIT Press. The podcast was produced by the Convergent Science Network (https://www.convergentsciencenetwork.org/). Context: The stability of social systems depends critically on realizing sustainable methods of “collaboration,” yet how and by which means collaboration is achieved is not clearly understood; neither are the conditions or processes that lead to its breakdown or failure. Collaboration can be understood as cooperation between agents toward mutually constructed goals. Part of the reason for our lack of understanding is that the phenomenon of collaboration is, by nature, a highly multidisciplinary problem, and effective research into its complexities has been difficult to achieve across the broad range of scientific and technical disciplines involved. The need for a fundamental understanding of collaboration, however, has become increasingly important. Not only does humankind demand answers as it attempts to address critical challenges at multiple scales (e.g., climate change, migration, enhanced automation, social and economic inequality), but ever-increasing technological and economic means of interconnecting people and societies are disrupting long-established, familiar patterns of how we interact. Radical technological changes that are ongoing have the potential to reshape collaboration in ways that are currently hard to predict or influence (e.g., by altering configurations in interaction, information creation, and modes of communication). On one hand, such changes could disrupt hitherto stable forms of collaboration by affecting critical communication channels and traditional roles, as can be observed in the rapidly changing patterns in governance, commerce, and social interaction. Conversely, technology could lead to the emergence of novel, successful forms of collaboration that deviate from traditional “hierarchical” architectures. Evidence of this can be seen in areas as diverse as highly automated manufacturing plants, the open science movement, collaborative software repositories, user-centered services, and the sharing of economy-based modes of organization. Without a fundamental understanding of the mechanisms, processes, and boundary conditions of collaboration, it is not possible to evaluate or predict which of these possible scenarios are sustainable or even plausible. The Forum “How Collaboration Arises and Why it Fails” (May 8–13, 2022, Location: Frankfurt am Main, Germany) Chairs: Andreas Roepstorff and Paul Verschure Program Advisory Committee: Jenna Bednar, Julia R. Lupp, Bhavani R. Rao , Andreas Roepstorff, Ferdinand von Siemens, and Paul Verschure

More posts

Timestamp

  • fast_forward00:00:00 - This is Paul for sure. I'm talking to Chao-Jing Wang, one of our speakers in the summer school.
  • fast_forward00:00:08 - And in your talk, you focus very much on the role of working memory,
  • fast_forward00:00:14 - sustained activity and cognition.
  • fast_forward00:00:16 - You started with a fairly strong
  • fast_forward00:00:18 - claim where you said that actually delay activity is the key to cognition.
  • fast_forward00:00:25 - So why do you think this is the key ingredient that we should worry about?
  • fast_forward00:00:33 - Right. I guess the general idea is that imagine that you cannot hold in your
  • fast_forward00:00:39 - mind anything in absence of direct sensory stimulation.
  • fast_forward00:00:44 - Then it seemed to me that most of the repertoire of behaviors is going to be
  • fast_forward00:00:49 - reduced to reflex all you can do is you know just,
  • fast_forward00:00:54 - reactively respond to stimuli right away otherwise you wait you forgot what
  • fast_forward00:01:01 - was the stimulus so you cannot really act accordingly and you become enslaved
  • fast_forward00:01:06 - to the external world whereas if you have this ability to hold something in
  • fast_forward00:01:11 - your mind even when the input is gone on,
  • fast_forward00:01:13 - then you are freed from the immediate stimulation and so that you can become
  • fast_forward00:01:20 - a lot more flexible, right?
  • fast_forward00:01:22 - You can, for example, wait and decide what you do about the stimulus and still
  • fast_forward00:01:28 - remembering what was the stimulus.
  • fast_forward00:01:31 - Right. Yeah. And then as an implementation of this, you were pointing very much to cortical circuits.
  • fast_forward00:01:38 - So what's so special then about cortical circuits that they can actually can
  • fast_forward00:01:42 - give you this kind of memory functions.
  • fast_forward00:01:45 - Right. So, in fact, persistent activity itself is probably more widespread than
  • fast_forward00:01:51 - working memory-related persistent activity. So, I want to distinguish those two things.
  • fast_forward00:01:56 - For example, even when you try to hold a gaze, your eyes are fixating on something.
  • fast_forward00:02:04 - That's maintained by persistent activity in the ocular motor system outside of the cortex.
  • fast_forward00:02:11 - But there's a very long history of studies pointing to the prefrontal cortex,
  • fast_forward00:02:20 - as a very important structure for working memory.
  • fast_forward00:02:23 - So that dates back to, I guess, 1920s and 30s, where people showed that if you
  • fast_forward00:02:30 - do lesion of the prefrontal cortex on an animal like a monkey,
  • fast_forward00:02:35 - we're not able to do delayed response tasks, which depend on working memory.
  • fast_forward00:02:39 - So from there on, I think, you know, there is a lot of studies showing that
  • fast_forward00:02:44 - prefrontal cortex is really important for, you know, working memory.
  • fast_forward00:02:50 - And that's why there's a lot of focus on cortex.
  • fast_forward00:02:56 - But then you were saying that it is the special characteristic of this dense
  • fast_forward00:03:02 - recurrence in these circuits that could sustain this memory function.
  • fast_forward00:03:08 - Is that correct? That's the idea, yeah. Okay, so then I could argue that actually
  • fast_forward00:03:13 - this kind of dense recurrence I would find throughout the cortex.
  • fast_forward00:03:17 - So why is then not my occipital cortex, that's usually more dedicated to vision,
  • fast_forward00:03:21 - actually a working memory system showing sustained activity?
  • fast_forward00:03:25 - What's then so special about these frontal areas?
  • fast_forward00:03:30 - That they can support working memory.
  • fast_forward00:03:34 - So I guess, empirically, we do not have enough data to answer your question explicitly.
  • fast_forward00:03:43 - But generally, it probably is a matter of degree, right?
  • fast_forward00:03:47 - So we know even in the primary sensory areas, the majority of synaptic inputs
  • fast_forward00:03:53 - are coming from within the local circuits.
  • fast_forward00:03:55 - But maybe there's just more in the brief on the cortex compared to sensory areas.
  • fast_forward00:04:04 - And what's interesting from the computational point of view is that you can
  • fast_forward00:04:08 - show that because of these feedback systems, your dynamics is nonlinear.
  • fast_forward00:04:13 - And in nonlinear systems, I think it's very important that just some graded
  • fast_forward00:04:19 - differences, some quantitative differences, for example, in the amount of recurrent connections.
  • fast_forward00:04:25 - Can lead to qualitatively different behaviors.
  • fast_forward00:04:29 - Like, you know, if your recurrent connections is below a threshold level,
  • fast_forward00:04:34 - you don't see persistent activity.
  • fast_forward00:04:35 - If it's above a threshold level, you start to see persistent activity.
  • fast_forward00:04:39 - So that may be my take about that. Just to add that it could also be there's
  • fast_forward00:04:44 - something else, such as neuromodulation, which may be somewhat different.
  • fast_forward00:04:48 - So, for example, dopamine modulation ventilation may be, say,
  • fast_forward00:04:53 - somewhat more prominent in the peripheral cortex than in early sensory systems,
  • fast_forward00:04:59 - for example, that could make a difference as well.
  • fast_forward00:05:02 - But I would say, again, it's a matter of quantitative differences leading to
  • fast_forward00:05:07 - qualitative differences.
  • fast_forward00:05:09 - Okay. But that's still a very much cortex-centered view.
  • fast_forward00:05:15 - So does it mean that you would be making the strong statement that okay working
  • fast_forward00:05:19 - memory can be fully realized by cortical circuits and does not depend on subcortical
  • fast_forward00:05:26 - activity beyond possibly some forms of neuromodulation so.
  • fast_forward00:05:31 - Yeah we don't know much about that i guess there's one thing certainly people think,
  • fast_forward00:05:38 - um for why subcortical structures might be important that is skating so you
  • fast_forward00:05:44 - don't want not any sensory stimulus coming into your brain to be stored in working memory, right?
  • fast_forward00:05:50 - So basically, you somehow have to gauge what really is behaviorally important
  • fast_forward00:05:54 - that you have to maintain internally, let's say, in the prefrontal cortex.
  • fast_forward00:06:00 - And that seems to depend on basal ganglia. There's this idea that,
  • fast_forward00:06:04 - you know, basal ganglia, for example, through the thalamus can gauge what signal
  • fast_forward00:06:10 - is important and what is not.
  • fast_forward00:06:11 - At least you know part of the gating function is dependent on basal ganglia okay yeah so,
  • fast_forward00:06:20 - It could also be that persistent activity itself may, in part,
  • fast_forward00:06:25 - depend on reverberation in a bigger loop involving the subcortical system.
  • fast_forward00:06:33 - But for that, people have speculated about that for a long time.
  • fast_forward00:06:37 - As far as I know, we do not yet have very strong evidence for that. Yeah.
  • fast_forward00:06:43 - So then the core ingredients of your model will be, let's say,
  • fast_forward00:06:48 - recurrently coupled excitatory circuits with, let's say, some inhibitory add-ons to get selectivity.
  • fast_forward00:06:57 - And this is, I think, some summary of a cortical circuit.
  • fast_forward00:07:02 - And this would then project downstream to, let's say, other subcortical structures
  • fast_forward00:07:08 - to trigger action, as in, for instance, your standard two-choice saccade-based tasks.
  • fast_forward00:07:15 - It would be like the choice aspects would happen in this cortical model,
  • fast_forward00:07:18 - and then the saccade would be triggered by, let's say, superior colliculus to
  • fast_forward00:07:22 - which this decision-making system would project.
  • fast_forward00:07:25 - So is that an assumption you really would like to insist on,
  • fast_forward00:07:30 - that these action elements are extra cortical, and that more the sensory-based
  • fast_forward00:07:35 - and the rule-based aspects are processed at the cortical level?
  • fast_forward00:07:44 - Well, that, I guess, depends. So in a way, saccade is special, right?
  • fast_forward00:07:51 - So if you talk about manual responses, then maybe parts of the motor cortex
  • fast_forward00:07:58 - are the command centers.
  • fast_forward00:08:01 - And so that would be kind of downstream system from decision circuits where
  • fast_forward00:08:10 - information are integrated, maybe a choice is produced.
  • fast_forward00:08:13 - So in that case, then even the, I think it's the part of the response generation
  • fast_forward00:08:20 - is occurring inside the cortex,
  • fast_forward00:08:23 - you know, always together with basal ganglia and some others like thalamus,
  • fast_forward00:08:28 - but in that case, you know, a lot of things may happen in the cortex as well. Okay.
  • fast_forward00:08:34 - So your, the approach, I mean, also given your background in physics,
  • fast_forward00:08:39 - you have used one of these tools that has been important to theoretical neuroscience
  • fast_forward00:08:44 - from physics of, let's say, attractor networks to analyze this model.
  • fast_forward00:08:50 - So why is the notion of an attractive network helping you to understand this
  • fast_forward00:08:55 - pre-protocol, these cortical dynamics?
  • fast_forward00:09:02 - You know sometimes i like to say especially to
  • fast_forward00:09:05 - experimentalists that the the uh the
  • fast_forward00:09:07 - word attractor networks seem to
  • fast_forward00:09:10 - have certain seem to provoke certain reactions in some people right so i would
  • fast_forward00:09:15 - sometimes start out uh by saying that uh attractor dynamics is not like a black
  • fast_forward00:09:23 - hole so it's not something that's very rigid that if you are in a tractor state
  • fast_forward00:09:28 - then it's like if you're You're sucked in.
  • fast_forward00:09:30 - You cannot get out. And it's a very rigid kind of thing.
  • fast_forward00:09:35 - Attracted states are simply relatively stable states.
  • fast_forward00:09:39 - That's all. It could even be something, you know, more than just a steady state.
  • fast_forward00:09:46 - For example, you could have chaotic attractors where you have a lot of temporal
  • fast_forward00:09:51 - dynamics going on inside that, you know, chaotic attracted state, right?
  • fast_forward00:09:56 - That's one thing. The other thing is that any inputs or neural modulation can
  • fast_forward00:10:02 - easily change, if you like, the landscape of attractive states.
  • fast_forward00:10:07 - So you can easily actually control,
  • fast_forward00:10:09 - manipulate, you know, the landscape of multiple attractive states.
  • fast_forward00:10:14 - So, you know, simply, as I was saying, you know, attractive states are simply
  • fast_forward00:10:19 - relatively stable states.
  • fast_forward00:10:21 - And if you want to describe mathematically what is a persistent activity, right?
  • fast_forward00:10:29 - Then it's natural to think about that as a relatively stable state that's all, right?
  • fast_forward00:10:36 - So as a result the concept of attractive states is very natural people still
  • fast_forward00:10:41 - debate about the actual dynamical structures,
  • fast_forward00:10:45 - of persistent activity that should be described well by attractive networks,
  • fast_forward00:10:52 - but that's I think a separate issue from a more general question whether,
  • fast_forward00:10:58 - attractor networks is the right framework or not to describe persistivity.
  • fast_forward00:11:03 - Because I could argue that in some sense any system that will show some persistence
  • fast_forward00:11:11 - of whatever spatial temporal scale you want to express this I can re-describe
  • fast_forward00:11:16 - in terms of an attractor network.
  • fast_forward00:11:18 - So the risk might then be that the framework of an attractor is so,
  • fast_forward00:11:23 - let's say, super powerful powerful, that it might not give you much leverage
  • fast_forward00:11:26 - with respect to a specific phenomenon.
  • fast_forward00:11:30 - So what's the leverage it has given you to really understand prefrontal cortex?
  • fast_forward00:11:36 - Sorry, I'm not sure I really understand the... Well, it's very simple.
  • fast_forward00:11:40 - In some sense, you could argue that if you look at the nervous system,
  • fast_forward00:11:44 - you will find at many different levels, you will find persistent states in some temporal window.
  • fast_forward00:11:51 - It can be a microsecond. It doesn't matter. Yeah.
  • fast_forward00:11:54 - In all those cases, I could step in with an attractor network formulation.
  • fast_forward00:11:58 - Ah, you see, it's an attractor because the system in some way is returning to
  • fast_forward00:12:02 - this state and then I define that state in some way.
  • fast_forward00:12:05 - So that means the attractor framework is super powerful.
  • fast_forward00:12:11 - You can describe anything with it you want, as long as there's some persistence
  • fast_forward00:12:14 - in the system, in some definition of its possible states.
  • fast_forward00:12:19 - I see. So what's special about, say, prefrontal cortex? Exactly.
  • fast_forward00:12:22 - Why does it give you leverage to understand prefrontal cortex?
  • fast_forward00:12:25 - I guess the important thing is that you have multiple attractor states at the same time.
  • fast_forward00:12:32 - If you think about if you want to design
  • fast_forward00:12:35 - a working memory system it basically is a system with multiple states so that
  • fast_forward00:12:41 - you can switch on and off between different states and that may not be so universal
  • fast_forward00:12:47 - in different systems for example you don't need and you don't want multiple.
  • fast_forward00:12:54 - States in not necessarily in early sensory systems for example And so even if
  • fast_forward00:13:01 - you have some persistency,
  • fast_forward00:13:02 - the question is whether you have this ability to go back and forth between many
  • fast_forward00:13:11 - different states with very brief input.
  • fast_forward00:13:16 - Does that make sense? So that seemed to be the distinction between what you
  • fast_forward00:13:22 - want for working memory system versus a sensory processing device.
  • fast_forward00:13:27 - Okay, so you're saying the framework of attractors gives you leverage in this case,
  • fast_forward00:13:32 - because actually you're looking at a system that can be in a large number of
  • fast_forward00:13:39 - possible states, and this is an efficient way to describe these. That's right.
  • fast_forward00:13:43 - And the persistency should be at the long time scale compared to the time constants
  • fast_forward00:13:49 - you have in the system, like milliseconds versus tens of milliseconds by physical time constant.
  • fast_forward00:13:57 - Let me also just briefly say that, as I was discussing in my lecture, that,
  • fast_forward00:14:06 - in fact, fast switches tend to be not the right conceptualization in the model I described.
  • fast_forward00:14:13 - So you also have very slow transients, too, and that's not just steady states.
  • fast_forward00:14:19 - And those slow transients, like a gradual ramping activity, turns out to be
  • fast_forward00:14:24 - a very good computational mean to, say, integrate information in decision-making.
  • fast_forward00:14:30 - So this is kind of an unusual type of attractive networks where you,
  • fast_forward00:14:36 - at the same time, have multiple stable states, But you also have this ability
  • fast_forward00:14:42 - to have very slow transients for time integration.
  • fast_forward00:14:48 - Okay, but do you see that as an intrinsic property of these circuits or as an
  • fast_forward00:14:56 - add-on that might be supported by a different substrate? Right.
  • fast_forward00:15:03 - So at least for the mechanism we found in the model, it's the same circuit.
  • fast_forward00:15:11 - So the same circuit, because the recurrent dynamics is mediated by,
  • fast_forward00:15:17 - you know, kind of slower cellular mechanism,
  • fast_forward00:15:21 - you can have both slow transients and multiple statist. Okay. Yeah.
  • fast_forward00:15:27 - Whether, you know, in the brain, those two things, actually,
  • fast_forward00:15:33 - there is evidence, I guess, at a single neuron level that in the brain,
  • fast_forward00:15:36 - those two things can happen at least in the same circuits, right?
  • fast_forward00:15:42 - Yes. And actually very commonly, you observe decision-related neural signals
  • fast_forward00:15:48 - in a decision task, and working memory-related signals in a working memory task,
  • fast_forward00:15:54 - in very different experiments done in different labs, but from the same code.
  • fast_forward00:16:00 - Adjusting, you have a shared mechanism. Okay.
  • fast_forward00:16:05 - Although you can never exclude that these might be interspersed circuits that
  • fast_forward00:16:10 - perform different functions, no? Right, in theory, sure.
  • fast_forward00:16:14 - But before we move into this issue of timing and decision-making,
  • fast_forward00:16:19 - another issue around attractor networks is that in order to be,
  • fast_forward00:16:23 - let's say, a certified and card-carrying member of the club of attractor networks,
  • fast_forward00:16:30 - you must satisfy certain minimal criteria.
  • fast_forward00:16:33 - Like when it's under perturbation, you must resort to the same attractor state, and so on.
  • fast_forward00:16:39 - So are you sure that these cortical, the frontal cortical network should look
  • fast_forward00:16:44 - at do satisfy all these conditions?
  • fast_forward00:16:52 - Well, I guess that's rather difficult to test, especially in behaving animals
  • fast_forward00:16:59 - or behaving primates, especially.
  • fast_forward00:17:02 - So I imagine that if people now are making efforts to develop maybe simpler
  • fast_forward00:17:10 - model systems like rodents,
  • fast_forward00:17:14 - if you can design a good, say, working memory or decision-making task,
  • fast_forward00:17:18 - it's probably more likely you
  • fast_forward00:17:20 - can really go down into microcircuit mechanisms and get
  • fast_forward00:17:24 - into more detailed information however you
  • fast_forward00:17:30 - can you know do certain experimental manipulations to test some model predictions
  • fast_forward00:17:36 - so for example as I showed that you know mentioned earlier that we found that
  • fast_forward00:17:43 - that reverberation should be slow.
  • fast_forward00:17:45 - In particular, it probably depends on a particular receptor,
  • fast_forward00:17:49 - NMDA receptors at the recurrent synopsis, right?
  • fast_forward00:17:53 - And we, in collaboration with Amy Anstin at Yale, actually we tested this idea
  • fast_forward00:17:57 - in behaving monkeys, you know, working memory task, using a technique called
  • fast_forward00:18:02 - ionophoresis to inject the drug locally.
  • fast_forward00:18:08 - Onto neurons you record from. So you see persistent activity in those neurons
  • fast_forward00:18:12 - in the prefrontal cortex of a behavior monkey.
  • fast_forward00:18:16 - And then when you apply the drug, the blocks and MDA receptors,
  • fast_forward00:18:20 - you actually see that process activity goes away.
  • fast_forward00:18:24 - That's, in my mind, a very direct confirmation that MDA receptors that are slow,
  • fast_forward00:18:31 - mediating slow repopulation, are critical for process activity.
  • fast_forward00:18:37 - Okay, yes, I see that. But then I could argue, well, but maybe you're blocking...
  • fast_forward00:18:42 - The NMDA receptors are the part of the thalamic projections.
  • fast_forward00:18:45 - Can you exclude that? No, you cannot.
  • fast_forward00:18:50 - That's right. But at least we can say it.
  • fast_forward00:18:58 - Are likely to be kind of intrinsic synapses
  • fast_forward00:19:01 - rather than you know synapses coming
  • fast_forward00:19:05 - from external stimulation because this is
  • fast_forward00:19:08 - during working memory during the delay when there's no direct
  • fast_forward00:19:11 - sensory stimulus right um so
  • fast_forward00:19:14 - to adjust your question you can do a different set of experiments but not with
  • fast_forward00:19:19 - monkey so with rodent you can do in vitro slices right and actually we have
  • fast_forward00:19:24 - done in collaboration with another group by Wenjing Gao in Philadelphia where
  • fast_forward00:19:31 - you do prefrontal cortical sizes directly.
  • fast_forward00:19:34 - You can look at, you know, NMDA or AMPA receptor immediately the transmission
  • fast_forward00:19:39 - between two cells neighboring neurons.
  • fast_forward00:19:43 - So there's really local connections between two neurons and show how much NMDA
  • fast_forward00:19:49 - you have at this very local synopsis.
  • fast_forward00:19:54 - In peripheral neurons, and you can compare that with the same kind of peer recordings
  • fast_forward00:19:59 - in the primary sensory system. You see a big difference.
  • fast_forward00:20:04 - Okay. That's also consistent. Right. So with these experiments...
  • fast_forward00:20:08 - So that would be really cortical, right? Local. Yes.
  • fast_forward00:20:11 - But with the antiparesis experiment, basically what you're saying is,
  • fast_forward00:20:14 - well, this makes it plausible possible that the memory state is implemented
  • fast_forward00:20:19 - by a recurrent or reverberating circuit.
  • fast_forward00:20:22 - But in some sense, what we still don't know is whether it really can be called an attractor or not.
  • fast_forward00:20:28 - So would you argue that, for instance, these micro-stimulation experiments that
  • fast_forward00:20:33 - have been performed in front of light fields that would sort of bias decision-making
  • fast_forward00:20:37 - of an animal, would you take that as corroborative evidence of having attractor states? Yes.
  • fast_forward00:20:44 - Or would that be the kind of way to get at that question? Well,
  • fast_forward00:20:48 - I think in order to go further along the question, I mean, what I suggest,
  • fast_forward00:20:54 - we should propose something very well defined that's an alternative to the attractive model, right?
  • fast_forward00:21:01 - And otherwise, it's hard to say, how do you prove or disprove the attractive network paradigm?
  • fast_forward00:21:08 - So I guess one alternative is slow transients.
  • fast_forward00:21:14 - Just because of some cellular or whatever, you know, biophysical process with
  • fast_forward00:21:20 - very slow time constant, right?
  • fast_forward00:21:22 - And so as a result, in response to transient stimulus, it just keeps going for
  • fast_forward00:21:28 - a long time on the timescale of many seconds, right?
  • fast_forward00:21:32 - So that's something we can try to, right, to contrast with a charted network.
  • fast_forward00:21:38 - And I'm sure slow processes are playing a role The question is whether it's
  • fast_forward00:21:45 - the workhorse, the main thing, right?
  • fast_forward00:21:48 - I do see a possible problem with a mechanism that's completely based on very slow process.
  • fast_forward00:21:56 - Because if you have a system with intrinsically very slow time constant.
  • fast_forward00:22:02 - As you know, you would have a hard time to change the system with briefing.
  • fast_forward00:22:10 - Because the time constant is so short and so long, You have to use very long
  • fast_forward00:22:15 - Z inputs to do anything with the system.
  • fast_forward00:22:18 - So switching on and off becomes a problem.
  • fast_forward00:22:21 - Right. Yeah. So maybe people can think about a clever way to really propose
  • fast_forward00:22:27 - very clearly defined alternatives, right? And then try to design. Right.
  • fast_forward00:22:34 - So what was very exciting about your proposal is that you also made the argument,
  • fast_forward00:22:39 - look, I can represent the memory state. I can represent the decision-making
  • fast_forward00:22:44 - by switching between my different detractors.
  • fast_forward00:22:46 - But on top of that, I can also capture the kind of ramping functions that have
  • fast_forward00:22:50 - to do with the reaction times as observed in prefrontal cortex, right?
  • fast_forward00:22:54 - Where you see that, okay, these neurons that seem to correlate with a decision
  • fast_forward00:22:57 - ramp more slowly when you have long reaction times and they ramp very fast when
  • fast_forward00:23:02 - you have a fast reaction time.
  • fast_forward00:23:04 - So how could that drop out of your model so easily?
  • fast_forward00:23:10 - What was the trick there? Why does that work?
  • fast_forward00:23:16 - Well, I guess the main thing is this idea of slow reverberation, I guess.
  • fast_forward00:23:21 - And that was a surprise to us.
  • fast_forward00:23:25 - You know, our priority, as I was saying, you know, working memory itself, in principle, right?
  • fast_forward00:23:31 - If you're an engineer, you think about how to design a working memory device,
  • fast_forward00:23:35 - you could say that can be done by faster switches.
  • fast_forward00:23:39 - But when you try to do that with neurons and the realistic synaptic connections,
  • fast_forward00:23:44 - you know, it turns out that the system is very unstable because of the strong feedback,
  • fast_forward00:23:52 - you know, machinery in the system.
  • fast_forward00:23:54 - And one way to solve that is to say positive feedback needs to be slow relative to negative feedback.
  • fast_forward00:24:03 - And that so was kind of forced on us, you know, on us.
  • fast_forward00:24:07 - So we say, you know, we have to have a working memory mechanism with
  • fast_forward00:24:10 - slow reverberation rather than very fast positive feedback
  • fast_forward00:24:13 - and from there it turns out um
  • fast_forward00:24:17 - you know uh it becomes easy so so this slow river version turns out to be exactly
  • fast_forward00:24:22 - what you need to get a slow ramping activity in decision that's i thought it's
  • fast_forward00:24:28 - quite nice yeah but now one one question i would have there is um.
  • fast_forward00:24:35 - In some sense in in this manipulation you could have
  • fast_forward00:24:38 - a confound that um what we are
  • fast_forward00:24:41 - what we are detecting what we are detecting is based
  • fast_forward00:24:44 - on moving dots essentially and
  • fast_forward00:24:47 - it's about the coherence of the moving dots that you
  • fast_forward00:24:50 - make your decisions so i could
  • fast_forward00:24:53 - argue look if i have these prefrontal neurons that
  • fast_forward00:24:56 - are sensitive to the moving dots then that
  • fast_forward00:24:59 - if i if they have an orientation tuning
  • fast_forward00:25:02 - then of course i'm driving them more effectively if
  • fast_forward00:25:05 - i have coherent movement in my scene than if i have if
  • fast_forward00:25:09 - i drive all these are incoherently so that basically means per unit
  • fast_forward00:25:11 - time they get less energy and therefore as a result
  • fast_forward00:25:14 - i'll be ramping up more slowly or faster you see and this then also exactly
  • fast_forward00:25:19 - correlates with the task condition so imagine i would change the task now that
  • fast_forward00:25:24 - the animal has to make a decision based on let's say um when when there's incoherent
  • fast_forward00:25:30 - movement it gets reward and it should ignore coherent movement so i sort of
  • fast_forward00:25:33 - i change the contingency,
  • fast_forward00:25:34 - would you predict the model would still work or would you have to add a new feature?
  • fast_forward00:25:39 - So what we added, we do need to add something in that case that is reward-dependent plasticity.
  • fast_forward00:25:46 - So in that case, you have to learn what are the potential outcomes from your choice options, right?
  • fast_forward00:25:55 - And that, we believe, is done through learning that depends on say, reward. Okay?
  • fast_forward00:26:03 - I think what you are pointing earlier at is the confine that maybe slow RAM
  • fast_forward00:26:11 - just corresponds to weaker inputs.
  • fast_forward00:26:17 - But in part it's true. So basically, you probably need to integrate more over
  • fast_forward00:26:22 - time if your input, your evidence is weaker.
  • fast_forward00:26:27 - But it's not just that, because, for example, we showed that if you have more
  • fast_forward00:26:33 - several options, more options to consider, then the reaction times are also slower.
  • fast_forward00:26:39 - And that in part is because you have this competition between your own pools
  • fast_forward00:26:44 - selected for say four or five options.
  • fast_forward00:26:48 - And that involves inhibition because it's a competition mediated by inhibition
  • fast_forward00:26:53 - that also slows down the ramping activity.
  • fast_forward00:26:56 - So you can also show nicely, you know, something that people see in psychology,
  • fast_forward00:27:00 - you know, the more options you have to consider, the slower the reaction time.
  • fast_forward00:27:05 - Okay. Yeah. Yes. so the other.
  • fast_forward00:27:09 - Aspect that I was curious about if you look at the model it's
  • fast_forward00:27:13 - also an issue we discussed earlier the interpretation of
  • fast_forward00:27:16 - this prefrontal cortical function also in the literature at
  • fast_forward00:27:19 - large it's in the end very much a labeled line kind
  • fast_forward00:27:22 - of system I have choice options the choice options depend on certain sensory
  • fast_forward00:27:26 - cues and if you want by magic they just come together in these units or in my
  • fast_forward00:27:31 - attractor network and now I can start to make decisions with them but you could
  • fast_forward00:27:37 - then of course pose the question, and sometimes it's a form of the symbol grounding problem,
  • fast_forward00:27:41 - okay where do these labeled lines come from and should I really assume that
  • fast_forward00:27:46 - these are labeled lines, that let's say your prefrontal cortex has all these
  • fast_forward00:27:49 - labeled lines projecting into it from other areas,
  • fast_forward00:27:53 - representing all possible cues you can ever encounter or you have ever encountered
  • fast_forward00:27:57 - in the world combined with all possible actions that you can ever trigger in
  • fast_forward00:28:01 - response to these cues, but that would be the labeled line view.
  • fast_forward00:28:04 - So do you think that that's a reasonable assumption, or do we have to get away from it in some way?
  • fast_forward00:28:12 - Right. So I guess you're totally right.
  • fast_forward00:28:17 - We have been focusing on certain elementary and fundamental,
  • fast_forward00:28:24 - machineries about working memory and decision-making without paying too much
  • fast_forward00:28:30 - attention to real-life stimuli, right?
  • fast_forward00:28:34 - And how do we really process real-life stimuli?
  • fast_forward00:28:36 - So I guess in my mind, to go forward in that direction, we have to understand
  • fast_forward00:28:44 - better how objects are recognized and represented in the brain.
  • fast_forward00:28:49 - And I could imagine, for example, in the inferior temporal cortex.
  • fast_forward00:28:56 - There we don't really understand yet how we recognize objects fully.
  • fast_forward00:29:02 - So it could be gilet l'ail, but it could be something more.
  • fast_forward00:29:07 - You know people talk about grandmother cells that's more
  • fast_forward00:29:11 - like data line I guess in your terminology
  • fast_forward00:29:14 - but or maybe something
  • fast_forward00:29:17 - more dynamic right we don't know yet what that is but to answer your question
  • fast_forward00:29:25 - I can envision that you know starting with the building blocks basically elementary
  • fast_forward00:29:30 - elementary machinery we kind of have some insights into
  • fast_forward00:29:35 - four working memory systems,
  • fast_forward00:29:37 - we can envision to connect that with something like a sensory,
  • fast_forward00:29:45 - visual system, including IT, for example, and see how the interaction between
  • fast_forward00:29:49 - a working memory system and the rest,
  • fast_forward00:29:52 - the posterior part of the visual system, together in a larger scale brain system
  • fast_forward00:29:58 - to carry out more realistic kind of representation and the working memory.
  • fast_forward00:30:06 - And I do think that's a major challenge in the field. How we have in Pakistan
  • fast_forward00:30:11 - mostly local circuits, right?
  • fast_forward00:30:13 - And we have to come up with a theoretical framework, new concepts perhaps perhaps
  • fast_forward00:30:21 - to understand large-scale brain system with interacting parts. Exactly.
  • fast_forward00:30:29 - So to go in that direction, if you look at the physiology on prefrontal cortex,
  • fast_forward00:30:35 - these classic experiments by Assad, Miller, and so on, in the end you look at
  • fast_forward00:30:39 - populations of neurons that actually have a very broad range of representations,
  • fast_forward00:30:44 - of cues, actions, and their combinations following different rules.
  • fast_forward00:30:48 - So, could you then imagine that maybe you have, let's say, more specialization
  • fast_forward00:30:53 - in these prefrontal circuits, that some neurons, let's say, contribute to cue information,
  • fast_forward00:30:57 - others to action information, and then in their interaction,
  • fast_forward00:31:01 - they will build up this, what you would call the attractor, representing this
  • fast_forward00:31:04 - key state in which you want to make your decisions.
  • fast_forward00:31:06 - Would that be a reasonable alternative? Yes.
  • fast_forward00:31:11 - Well, actually, so we have done more recent work with Stefano Fusi and Mattia, his graduate student.
  • fast_forward00:31:22 - That work suggests, this is a computational work, suggests that,
  • fast_forward00:31:27 - in fact, the neuron signals in the peripheral cortex probably should have a
  • fast_forward00:31:34 - lot of mixed selectivity.
  • fast_forward00:31:36 - So a given neuron would be, you know, activated to different degrees by a combination
  • fast_forward00:31:43 - of many different things, including, you know, sensory stimuli,
  • fast_forward00:31:46 - internal representation of behavior rules,
  • fast_forward00:31:49 - and maybe some control signals altogether.
  • fast_forward00:31:52 - Together so i my view is
  • fast_forward00:31:55 - that neurons are probably not dedicated to one thing only especially in the
  • fast_forward00:32:00 - driven accordance you know maybe in contrast to early sensory systems and there
  • fast_forward00:32:05 - it's very likely that you know there's a lot of multi a lot of mixed selectivity
  • fast_forward00:32:11 - and so that you know neuron Neuron groups,
  • fast_forward00:32:14 - if you like, coding things according to a combinatorial code,
  • fast_forward00:32:21 - that will be maybe a very good way to be able to combine information, right?
  • fast_forward00:32:32 - Another thing that's special about peripheral cortex is that it gets inputs
  • fast_forward00:32:36 - for many, many different areas, right?
  • fast_forward00:32:38 - And that also speaks to the fact that neural inputs also are kind of coming
  • fast_forward00:32:45 - from a lot of convergent-divergent pathways. Right.
  • fast_forward00:32:49 - So then on top of that, where do you see these models go?
  • fast_forward00:32:54 - What's the next challenge for you in this model? Okay.
  • fast_forward00:33:05 - So I guess briefly I can see several directions.
  • fast_forward00:33:11 - Number one is to see if, you know, the kind of models we build can be kind of,
  • fast_forward00:33:20 - and those are building blocks of cognition.
  • fast_forward00:33:22 - I like to call this, right, local circuit mechanisms and building blocks for cognition.
  • fast_forward00:33:27 - So the question is, can you extend this kind of approach, at least,
  • fast_forward00:33:32 - to more complex behaviors, more complex functions,
  • fast_forward00:33:36 - such as task switching or rule-based behavior, right?
  • fast_forward00:33:46 - So that's one. Number two, I just mentioned earlier, you know,
  • fast_forward00:33:50 - can we develop theory and computational mechanisms and the principles for large-scale
  • fast_forward00:33:57 - brain systems rather than just local circuits?
  • fast_forward00:34:01 - And third is real-life situation.
  • fast_forward00:34:03 - That definitely will challenge our kind of models in a very big way.
  • fast_forward00:34:10 - So we want to see if the insights from those kind of models are really useful
  • fast_forward00:34:18 - for our behavior in the face of a natural kind of environment.
  • fast_forward00:34:25 - Environment yeah so then what what
  • fast_forward00:34:29 - i also realized is that recently you have been become more interested
  • fast_forward00:34:32 - in let's say brain rhythms and brain oscillations and
  • fast_forward00:34:36 - you wrote a rather impressive review article on on these oscillations reason
  • fast_forward00:34:41 - why why how is this related to the attraction network so this is really a new
  • fast_forward00:34:45 - chapter so this is a very interesting topic i think more and more people are
  • fast_forward00:34:52 - interested in this really,
  • fast_forward00:34:55 - very active field,
  • fast_forward00:34:58 - you know.
  • fast_forward00:35:01 - People from very different disciplines, you know, like systems and neuroscientists,
  • fast_forward00:35:06 - as well as actually many people in the clinical field, people feel like this
  • fast_forward00:35:13 - synchrony, neuron synchrony, might be a way to look at, for example,
  • fast_forward00:35:19 - interactions between brain systems, right?
  • fast_forward00:35:21 - So, you know, related to large-scale dynamics type of issues.
  • fast_forward00:35:26 - We have worked a lot on, you know, or the mechanisms of synchrony and possible functions.
  • fast_forward00:35:36 - You know, this emerges naturally in recurrent networks, basically. Okay.
  • fast_forward00:35:43 - But it's still rather controversial, especially in terms of functional implications.
  • fast_forward00:35:51 - In part, I think it's because you do see often evidence of synchrony and oscillations
  • fast_forward00:36:01 - with measurements like EG or local field potential.
  • fast_forward00:36:06 - On the other hand, single neurons
  • fast_forward00:36:08 - are very stochastic. Even single neuron populations are very variable.
  • fast_forward00:36:13 - Okay. And so those two things don't seem to, you know, fit together.
  • fast_forward00:36:18 - Okay. That's, I think, part of the reason people feel like, you know,
  • fast_forward00:36:22 - it's hard to understand how do you explain those two things in the same framework.
  • fast_forward00:36:28 - And also, if the neuron operation is very stochastic,
  • fast_forward00:36:33 - then how much, right, in terms of degree, synchrony is really there on top of the stochasticity?
  • fast_forward00:36:43 - And if it's not big, right, if you have some measurement about synchrony,
  • fast_forward00:36:48 - and if it's still a few percent of the signal you see, why it should be the main focus?
  • fast_forward00:36:57 - Okay, I'm raising these questions without knowing the exact answers to those questions.
  • fast_forward00:37:01 - So, you know, in this review, for example, I try to discuss how you can reconcile these views.
  • fast_forward00:37:08 - Okay, and the general take, I guess we don't definitely know is the final answer.
  • fast_forward00:37:13 - But the general take is that maybe it's better to think about the role of synchrony
  • fast_forward00:37:19 - in a framework, in terms of neuron correlations, which we know are very important, right?
  • fast_forward00:37:27 - Neuron correlations in time, for example, are important for plasticity,
  • fast_forward00:37:32 - spike time-independent plasticity, right?
  • fast_forward00:37:35 - It may be even important to generate a stochasticity.
  • fast_forward00:37:39 - You know, it sounds like a bit of paradox, but we know that because a given
  • fast_forward00:37:43 - neuron receives a lot of inputs, you can average out noise.
  • fast_forward00:37:48 - And if, on the other hand, neurons are correlated weakly, they cannot average out noise.
  • fast_forward00:37:55 - And so maybe correlation itself is important to generate a stochasticity.
  • fast_forward00:38:02 - And that in turn, of course, can be functionally very important.
  • fast_forward00:38:06 - So to understand the different aspects of dynamics in the recurrent network
  • fast_forward00:38:13 - is a big challenge. So I guess that's a big reason that we are interested in signaling.
  • fast_forward00:38:18 - Very good. So then to finish up, two questions.
  • fast_forward00:38:22 - So coming from physics, going to neuroscience, attracting networks,
  • fast_forward00:38:29 - and actually now moving towards brain oscillations, large-scale understanding of the brain.
  • fast_forward00:38:36 - If you would have to stipulate a law that we should all follow in studying the
  • fast_forward00:38:40 - brain, so what would be the Shao-Yin-Guang law?
  • fast_forward00:38:46 - That's a tough one.
  • fast_forward00:38:53 - Well, let me phrase it this way.
  • fast_forward00:38:58 - I do think that trying to understand neural circuits of cognition is really
  • fast_forward00:39:05 - a very exciting, challenging thing.
  • fast_forward00:39:10 - And that would help to unify cognitive sciences and quote-unquote hardcore neurobiology,
  • fast_forward00:39:16 - which right now is still kind of separated.
  • fast_forward00:39:19 - And from what we learned, at least, is the key is slow reverberation balanced by inhibition.
  • fast_forward00:39:26 - And so again, quantitative differences can give you surprisingly new qualitatively
  • fast_forward00:39:34 - different functions and computations.
  • fast_forward00:39:37 - And that'll be, I mean, it's probably known from the theory of dynamical systems field,
  • fast_forward00:39:43 - but it should be a very big part of our efforts to understand cognitive circuits.
  • fast_forward00:39:54 - Okay, so judging one law is slow reverberation does the trick.
  • fast_forward00:39:58 - That's right. Very good. What time do you mean?
  • fast_forward00:40:02 - Okay. And then my last question is, if so, I'm going to go visit you five years from now at Yale.
  • fast_forward00:40:09 - I'm going to ask you then five years from now, like, look, five years back,
  • fast_forward00:40:13 - you gave me this one prediction that you would have believed in, so how did it pan out?
  • fast_forward00:40:16 - Was it false or true? what is one prediction you'll be willing to stick your neck out for today?
  • fast_forward00:40:28 - Well, I guess still slow recuperation. There are many other smaller predictions
  • fast_forward00:40:34 - that came out of the model.
  • fast_forward00:40:37 - You know, just one more example.
  • fast_forward00:40:41 - This is a very specific example, but I think it's quite impressive.
  • fast_forward00:40:46 - If it turns out to be correct.
  • fast_forward00:40:48 - That is, you find some scale invariance of reaction times.
  • fast_forward00:40:54 - That's at the behavioral level. It's a psychological law, really.
  • fast_forward00:40:58 - Very quantitative, very beautiful law in psychology that you can explain with
  • fast_forward00:41:04 - a neuron-circuit model mechanism in terms of stochastic neuron dynamics, right?
  • fast_forward00:41:10 - And, you know, that can be proved using neurophysiology.
  • fast_forward00:41:16 - I thought that would be really quite important, really, to relate what you see
  • fast_forward00:41:22 - in neurons in the recurrent circuit and what you see at the behavioral level.
  • fast_forward00:41:27 - Yeah, but that prediction has already come out, so I want a new one I can come
  • fast_forward00:41:31 - and hassle you with five years from now.
  • fast_forward00:41:36 - That has not been done yet. Of course. The thing is, I want you to now take
  • fast_forward00:41:42 - this risk that you might be wrong.
  • fast_forward00:41:46 - So.
  • fast_forward00:41:58 - I guess, you know, slow reverberation. But I guess how to prove that's wrong
  • fast_forward00:42:02 - is... It's a safe prediction.
  • fast_forward00:42:08 - Right. You know, we also want to have a big one, right? Or just not just any...
  • fast_forward00:42:15 - Whatever you feel comfortable with.
  • fast_forward00:42:16 - How about the following? Why don't you put some boundaries on slow?
  • fast_forward00:42:20 - So in what range of frequencies are we talking about for slow reverberation?
  • fast_forward00:42:28 - Yeah, so that, I mean, actually opened up a whole can of worms or opened a big
  • fast_forward00:42:34 - door because, you know, when you make decisions, you can integrate information over many timescales.
  • fast_forward00:42:41 - So, right, so in the brain, supposedly there are machineries that allow you
  • fast_forward00:42:46 - to integrate information, right?
  • fast_forward00:42:49 - So you can talk about integration over many seconds or even minutes.
  • fast_forward00:42:53 - So we don't know on that timescale what really is going on in the brain.
  • fast_forward00:43:00 - Let's see, what prediction that can be proven to be raw?
  • fast_forward00:43:06 - Let's make that very specific. I guess
  • fast_forward00:43:09 - I bet NMDA receptors and recurrent synapses inside cognitive-type neural cortical
  • fast_forward00:43:19 - circuits are the key for slow integration on a timescale of a second or so. Perfect.
  • fast_forward00:43:28 - Chai-Hsing Wang, thank you very much for this conversation.
  • fast_forward00:43:31 - Thank you. It's a pleasure.

Be the first to leave a comment

Leave a comment

Your email address will not be published. Required fields are marked *

Convergent PozitivLinia Logo

Exploring the convergence of neuroscience, robotics, and AI through conversations with leading researchers since 2010.

A project of the Convergent Science Network Foundation.

© CSN Podcasts. Developed by IMCreativeWEBC

0%

Login to enjoy full advantages

Please login or subscribe to continue.

Go Premium!

Enjoy the full advantage of the premium access.

Stop following

Unfollow Cancel

Cancel subscription

Are you sure you want to cancel your subscription? You will lose your Premium access and stored playlists.

Go back Confirm cancellation