Well, Gustavo, I've been reading his work since I was an undergraduate.

 I think of my Ph.D.

 student and a great expert in network connectivity in the brain where he was to apply this knowledge of neuroscience and physics together.

 So he's currently full professor catedrático at the University of Pompeu Fabra in Barcelona, where he's based, and he also runs the Center for Brain Cognition there and is an accredited professor and has had an EOC.

 Advance Grant is now part of the new Synergy Grant in neuroscience to be awarded in Spain and to work on basically stroke recovery in networks and in stroke recovery.

 And I was in Hamburg yesterday, and I just found out that Gustavo is a member of the Science Academy of Hamburg.

 So the list is very long.

 There also worked with Siemens and has an award from Siemens in terms of innovation for patents as well when we were talking about this idea.

 So yeah, and for those of you that know e-brains and so the Human Brain Project legacy and Gustavo was part of the Human Brain project is now a member of the scientific committee of e-brains, which there are two people per country.

 So I think it's pretty fair to say that, you know, he's not the not maybe.

 Well, we don't know.

 I would say the top neuroscientist in Spain at the moment.

 Yeah.

 So we're very happy to have him.

 So over to you. Thanks.

 Yeah.

 Thank you so much for the exaggerated and kind introduction, Brian.

 And I think especially for for inviting me and for having the possibility to present part of my interest nowadays from the scientific point of view.

 So the title of my talk is a little bit provocative as you realize the thermodynamics of mine.

 I mean, I hope that I think I have never found a paper on thermodynamics and neuroscience before.

 So from that point of view, it seems to be really a little bit provocative.

 But at the end of the day, you will see that the content is not provocative at all.

 As I mean, it's a very, very, very robust framework, very natural and very intuitive framework that we think can be used in neuroscience for answering some particular questions.

 And this is how I will start it just motivating the questions.

 So what we want to achieve, which kind of question we want to answer in this cognitive neuroscience field, the main question that we are trying to solve is a about the brain dynamics and who is running the brain dynamics?

 Who is orchestrating the brain dynamics?

 And this is what we call more technically the data organization in the brain dynamics.

 So just to give you a feeling, we want to know very, very fundamental things.

 For example, if I am doing nothing, what we call our data resting state, how is my hierarchical organization at the spatial temporal level?

 So they are all regions equally important.

 So we have a kind of democracy, we have a flat organization or there is a kind of in the other extreme dictatorship, there are some regions we are really running the shows and giving and driving.

 The rest of the brain is one question And then how would this theoretical organization change and differentiate to other, for example, cognition?

 So instead of doing nothing, I start to do a working memory task or decision making social task, whatever.

 So is something changing or not?

 I mean, from the hierarchical point of view and of course in different brain states, if I am sleeping, if I am under anesthesia, if I take drugs or in the biomedical context, if I am healthy, or I have some kind of neuropsychiatric disease, is this Iraqi colonization reflecting that?

 And if it is reflecting that is perhaps the causes of my disease or not?

 And if I know how, how is the right hierarchical organization and how is the brain, the irregular indecision in a particular disease perhaps, that give me some hints to not only predict and and and make some diagnosis on that, but even to design some possible therapies.

 So the the most natural way of starting to talk about the hierarchy actually is nothing new.

 I mean, people in artificial intelligence, in computer science, they know that it seems well, not centuries, but decades at least.

 And is it?

 Well, then just do it and measure your functionality and direction your interactions because you are interesting dynamics.

 So it should be a special temporal interactions.

 Just measure with your favorite measure of function and interactions and then put that in a graph and do it.

 You don't kill analysis, nothing else.

 It's nothing to to complex.

 I mean, all the tools are there.

 The problem is that's the type of data that we have in neuroscience are not always allowing this thesis.

 Try forward, detect the solution sometime is possible and I will show in the next slide.

 One example that is this is what they call the the detect this the most simple solution.

 For example, you can use Granger causality.

 We were talking about that at a couple of minutes ago and undefined your functional interaction in term of Granger causality and then you build your graphs under all the different situation and they start to analyze.

 Look, I mean in this conditions, that particular region is really running the culture.

 So what the whole talk is about, what we can do if we don't want to measure directly the functional interaction because we don't have enough data, that is the main problem.

 And we are using the three coming from the thermodynamics, which is sounds very technical, but you will see it's very, very simple.

 It's the breaking of the detail balance.

 So we know in the thermodynamic.

 But I will describe that with some details that in the talk that when you break the symmetry in the interactions, meaning if you break the symmetries, you start to build up hierarchical organization, then this is what we call the breaking of the detail balance.

 And that has consequences that could be reflected in other measurements which are not the functional interaction.

 For example, one very well-known measure is the arrow of time, and that is what they will use just by concentrated on the arrow of time of brain signals.

 Indirectly, I can say something about the hierarchy organization and that is fantastic because as we will see, the collective decision of that of time is a much more simple business to measure with the with a low amount, a small amount of data.

 And that is the talk.

 I mean, basically to to see how we can measure the theoretical or any session indirectly through this feature of the done while in time, namely that what we call out of time or non reversibility or non-equilibrium.

 So we we practically equalize all these words out of time, not of activity.

 The non-equilibrium and what is important for us is, you know, to get organization.

 I don't care if the brain is in article by like because I physicist but but I theoretical immunoassay I don't care if that is a narrow of time or it's just Barcelona with it but they care about the target organization and that is what we do.

 So in the first part that I, I will do this in a model three way and then at the end of the talk, I will say a couple of what about that we can use now these new measurements instead of the classical measure like function are going activity to job correlations or even sophisticated measure in cases that we can use that as I mentioned, greater causality before to construct a model.

 And then we will have some causal mechanistic explanation of this particular type of hierarchical organization.

 So that is the motivation.

 I'm I'm glad of that though, first very rapidly, because if you are in neuroscience, I mean, people were interested in the local organization, the whole brain level also since many decades.

 One typical example that we know is the standard.

 Can that together with the champion sans your use, it is this concept from bars that was a global workspace idea and it's very well reflected in this cartoon.

 I mean where would you see all these nodes, which are the brain regions and and the imagine it still strange regional hierarchy you know organize it where you have very physical brain regions probably more associated with sensory information processing and some central regions which are this global workspace which they are in the central circle which are regular 18 hour driving at orchestrate, orchestrating the the other regions.

 And in fact, as you know, they use this theory as an example.

 It's not necessarily the the main use of that for plane in a different brain instead like consciousness.

 So if the global workspace those of you the region has the top of the Iraqi allow the ignition of information so the transfer of information from the periphery to another part of the brain to another part the in the in the periphery than they claim, then there are some evidences that is associated with what we call consciousness.

 And if you are not allowing this this ignition that you see, they are in this read the example called 60 on the right called subliminal that you see are some activation but is not enough to to to to a start to this huge integration that you see for example in the green curve.

 Okay as I said, the most simple and direct way is really to measure functional interaction.

 We took an example where we were able to do that because we had a lot of data and that is the case of the Human Connection Project, as you know, is a large collection of a thousand people, a thousand participants, many different condition.

 Neuroimaging is if I mainly the result is mean, but I will concentrate only on them if MRI today and there are many conditions, resident state and many different cognitive condition that you see here.

 The Tauber like working memory, socio relational task model task language gambling. Imagine at this.

 And then if you take your measure of Granger causality, I took a transfer entropy normalize it with silvergate blah blah blah.

 I don't go into the details.

 Then you can measure directly the interactions and then you select who are at the top of the hierarchy.

 And those are the red regions in the top renderings.

 In each different conditions, you see that there are different regions running the Joe They are the I don't remember, I think I took a they are like five or 10% of the top regions and that is and do have more if you do.

 Now the intersection is a little bit more sophisticated at the intersection, but there is just roughly to give you a flavor is and it kind of intersection of all these red regions all the top regions in in different cognitive situations than what you get is what you see in B in this rendering this yellow and orange regions which are basically quantification of the global workers space.

 What the regions with and all possible cognitive demands are always running the shows now and the folk would be associated with that.

 So from that point of view, in that case, because we have so many and so high quality data, remember not only the large number of participants, but also two sessions, in some cases they are four session of resting state, long sessions, 22 minutes, and with the fabulous year of 0.7 seconds.

 So it's fantastic data.

 So with that you are you managed to really calculate the convention causality.

 Go now to another context which is also equally relevant psychiatric application, biomedical obligation.

 You are lucky if you get the resting state of 7 minutes with the bacteria from the hospital, 2 seconds and 20 subjects.

 You cannot calculate anything down.

 So therefore we were motivated and I want to apply all these ideas to to psychiatry.

 So I am obliged to develop a framework, a theoretical framework that allows me with this low amount of data to characterize and this is what we call that the more dynamic state of mind and now I will go a little bit more into the details.

 Before that, I suddenly started to have a result.

 Originally, I am coming from quantum mechanics, I mean actually from electrodynamics, quantum mechanics and a I and I was always even nowadays very interested in history of science and what you see there is the the, the image of a very well-known scientist, Edwin Schrodinger, a Nobel Prize in quantum mechanics.

 And of course, I mean he he had he ran very beautiful and very relevant ideas in quantum mechanics.

 And I got a question.

 For example, I was not aware that he was working also at the end of his life in in life science.

 And he had a great idea that he wrote in this book what this life.

 And it's a tiny book, very beautiful, written.

 And there he puts an idea, namely that we should use non-equilibrium thermodynamics in the context of biology because he associate basically life with non-equilibrium.

 But that's open a whole field in system biology.

 So at the molecular level where they they started to apply really those ideas of non-equilibrium thermodynamics in protein synthesis and so on, and continue even nowadays.

 So what I am doing today here is to try to apply exactly the same kind of ideas but in the macroscopic context.

 So I am not applying this to molecules, but I will apply to the whole way that is the joy attached to this claim that the idea is not coming from me.

 It's coming from him. And so here is a cartoon session of this idea now with a little bit more of that there.

 What you see on the top is a brain with just a cartoon.

 I said four regions.

 Of course, we will take hundred of regions most of the time.

 And then you if you study the the I introduction, assuming that we can measure that and if we know that the functional interaction are absolutely symmetrical, then automatically we know that the system is nonhierarchical.

 This is a democratic flat system.

 They have all the same problem.

 There is no a leader there.

 A From the thermodynamic point of view, this is called detail balance, this symmetry in the information flow.

 And then we know automatically two consequences.

 One that's that is no entropy production.

 And second, which is actually a theology of the first, that is an arrow of time.

 So that is that there is no out of time.

 Sorry, that is reversibility.

 If you go now to the bottom part of the slide and then you break the order, assumed that you are breaking the symmetry that you start to generate those rings like in the global workers base.

 So some of the central guys are more important because they're they could be driving the other guys.

 And then we know that if we break the symmetry in the functional interaction we have, yet our key and we have a breaking of the little violence and then exactly the control that we assumed before, we have non-equilibrium and that means that we have production entropy different from zero and we have non reversibility.

 Another of that and that is the good new because the production entropy is something that is very beautiful.

 But we can also not measure per say that paper is measured in that even from my lab.

 But we do dirty tricks that I am ashamed to talk about that today.

 We reduce the space, I mean, and using technique.

 And so but these are high dimensional systems.

 So to to estimate probability and entropy in the high dimensional system, forget it. But the idea but as I say, they were of that is the arrow of time.

 And this is what we measure.

 We just look, if there is a distinction in the in the time evolution of our but in signals, most of the time could be bold signals.

 I will show some example with the local field but then Jensen so that you get the feeling that it's not depending on the other measurement.

 So this is just the Hollywood version of the concept of not reversibility and reversibility.

 Just to give you a feeling that what you see in these two movies on the left is the non reversible system.

 Non-Equilibrium So a glass of wine which is destroyed by the impinging bullets that you see the for one movie and the same movie, but backwards and it's very clear what is for What about this fog one.

 Yeah.

 A On the other side you have a typical classical reversible system colliding consisting of volleyballs that would be perfectly elastic is perfectly reversible.

 In fact, what you see is the moving forward backward.

 And you cannot distinguish, you cannot say which one is the reality.

 Both that, okay, the problem is that it's not always the case.

 That is so easy to distinguish reversibility from the possibility.

 And this is the reason why I put here on the left.

 Actually, I have to confess I am a fan of Christopher Nolan, but Tenet is the worst movie, of course.

 But for me it was a good example because are people traveling forward and backward in time producing entropy or or the increase in entropy?

 You see in this excerpt of the movie that there are some people there, for example, traveling forward and backwards.

 It's very clear.

 It's like the glass of wine is is absolutely clear.

 What you see on the right.

 It is a tricky system.

 It's a physical system. It's not important in the detail.

 It is an expensive too, but that by construction, just skiing in that nature of physics paper, it generates a system that he knows because he simulated is by definition non reversible, is in non-equilibrium.

 And then you measure something, it's the magnetization and you see how the magnetization evolve in time is one of this kind of actually, I run the simulation, I run the simulation on the proposed budget game and I don't remember which one is the forward on the backward ration.

 So one caller is for when I'm the other college smuggler, I don't know and I cannot recognize.

 So meaning, even if I know that I should be able to recognize is not this is not intuitive A and the first year since kicking with a fabulous idea so basically this is assuming all this concept of out of time entropy in an non-equilibrium or so on.

 But the important part of this slide is the the the panel on the on the right.

 What this is reflecting is the in some cases it's very easy to distinguish like in the top four from backward A you don't need nothing.

 I mean you just need your eyes and that is in other cases like in this physical example of you are seeing, she is difficult, but we know how to distinguish things in difficult cases, especially nowadays.

 It is actually the field of of machine learning.

 A And therefore the idea was to use a machine learning network that we call on purpose.

 Then it but means a temporal evolution deep learning network in order to distinguish in the case of the brain, which is not like the case of the movie of Tenet it or the class of one, but more the difficult case and to use this machine learning technique in order to just distinguish the direction of the time to determine if there is a matter of time or not.

 So we apply this in all possible contexts.

 One can apply this to at the whole brain level, at the network level, at the local level, with different type of signals, different type of species.

 Of course, at the beginning they just concentrate on ball signals, neuroimaging in humans and the different conditions.

 And what you see in the middle panel is just in one dimension so that we can plot something is just your ball signal as you measured it in one region.

 A And then by hand you generate the backward motion, you flip it. Yes.

 And then you know which one you flip it and which one is the real measurement.

 And then you can train a deep learning classifier black box in a supervised way because you know who is or whose role or who is of output.

 And then see if you can separate that than using all the tricks of machine learning in a cross-validation, blah blah, blah blah.

 You check the performance if you see that the performance is high.

 So you can distinguish a signal that you have never seen before, that means that that is an arrow of time that means that the system is not reversible.

 That means that this is a non-equilibrium and the most important consequence is to see them easier to kick out and you have a quantification.

 So the performance of this network quantified the level of hierarchy.

 On the contrary, if it is low, you have more reversibility, more akin, leave, less less theoretical organization.

 But as I said, you can use that in different context.

 I am not going into the details.

 What you see on the right as a whole brain level network, labor or local level.

 I think I will give you an example of the of the two extremes global and local level.

 So this is one case again, HDP, so that thousand people neuroimaging here, but the resting state and all the different conditions I concentrate on the on the left panel, it's the important one.

 And you see on the x axis, all the conditions rest in first and then all the cognitive task.

 And on the y axis you see for the cross-validation set, therefore you see more or less hundred dots.

 I trained it the ten and with the more or less 900 participants and then a 200 random participant for cross-validation.

 And you check the the label of non CBT as a performance of the net for each single subset.

 And what you see is somehow expected they are or the reading increase in order of larger organization resting is the less dark color, but this hierarchical is not zero.

 You see, it's pretty guy.

 It's of one five, five, whatever.

 So there is a radical realization in your brain, even when you are doing nothing, which is not cytology, because we knew about the existence of networks.

 And so but it's nice to quantify that.

 And then what is expected, of course, that when you start to use your brain for doing something, of course you break even more the symmetry, even more the functioning director, because you need to promote the cross-talk very specifically between different members of your brains and therefore the organization is increasing.

 The beauty is that you are quantifying this and then you can see that there are different target organization in the social task compared to whatever emotion.

 And this is totally different.

 I don't comment on the other plots.

 This is just to give very rapidly to give you a flavor of biomedical application.

 This it is also a public dataset from UCLA, a It's very small, but what you have to see on the top left is exactly the same as we have seen before.

 But now for four groups of participants in resting state, the first group, the Gray is Healthy Group control, and then the other ones which are significantly below the the control group.

 So they have less hierarchical organization under the same condition reasons they are schizophrenic, bipolar and attentional deficit.

 And then you can go into the more specific description of that.

 This is a global measure, of course, of out of time.

 If you start to apply exactly the same philosophy at the local brain area level and then you rendered the differences is what you see.

 For example, at the bottom, just the difference between bipolar and controls, between attentional deficit and controls or schizophrenia and controls.

 And then you can start to interpret in which way that you darker organization change in all of them.

 I mean, in all of them globally goes down is flatter but in different ways.

 For example, I don't want to go into the details.

 You see, schizophrenia in general is going down all the regions in attentional deficit is mainly somatosensory or occipital areas going down, which makes sense.

 And one can interpret that that because they are more driven by the vitamin and therefore that then you don't get the gaze and things like that.

 Okay.

 Just very rapidly to give you a flavor, we can simplify this things and instead of using machine learning, we can use a measure which is even a much simpler to compute exactly the same concept out of time.

 The idea is following on the top line.

 You see how to calculate that for a pair of regions.

 And then in the bottom part is you can do this for all pairs in a matrix and then get the exact information at the whole brain level.

 The idea is very simple.

 Don't don't see that slide is too complex.

 I realize you measure the correlation but in a shifted way.

 So you break the symmetry in time on purpose of the duration of the two signals.

 So instead of calculating the correlation, as always you do the shifted correlations.

 Usually you move one to, I mean, some methodology for designing the optimal shifting, and then you do this with the forward version and get the number and then you do the backward maps.

 You and do exactly the same.

 If the system has an arrow of time, that will be different.

 So if you do, the difference of the two shifted correlation, this is what we call inside out for some strange reason, but it's a number and that we characterize the label of asymmetry in this in this case.

 And as I said, you can do that for all possible person that you have the whole brand label.

 Just again, another example in this case, another species, they are monkeys, another type of measurements.

 They are a and actually intra cortical EEG.

 So phase cortical EEG is all high quality electrical signals for monkeys.

 A difference in this case is different brain isolate conditions.

 They have manipulation.

 Whether you see the difference between awake and asleep or awake and the action of different anesthetic all under poverty.

 And what you see on the right path in all these books, blood is always the same is that measure of non reversibility meaning of hierarchy in all this condition and what you see is in conscious state, so in awake is always higher than asleep or higher than in unconscious state, like in under the effect of anesthetic or like propofol and D or ketamine and empty.

 And then when the effect of the anesthetic on disappear is the third part of the of the blood that you start to recover and the organization you start to build up again, there is only one exception, which is ketamine is going in the other way.

 And you can interpret that.

 The state reason why ketamine is not only an anesthetic but is nowadays also use it as a fun drug and is increasing the hierarchy in a funny way.

 This is a biomedical application of the same framework.

 These are data from years and berries.

 Coma patients controls minimal consciousness and deep coma, and you go in the right direction.

 A Of course controls have a certain level of about you not to give you any session a little bit less in minimal consciousness and much less significantly, much less in in the deep learning case shows now that we have a good measure that characterized the the hierarchical organization in let's see if we can explain that mechanistically.

 So what this means with this is a model for the measurement quantification and we can try now to build a model which explain exactly that measure.

 So the the get out of key to the quantification of the Iraqi organization.

 And so the the sense of the model I think is relatively well-known nowadays.

 But just to cartoon, is that what we call whole brain model and nowadays is mainly the integration of the coupling coming from the anatomy.

 For example, in humans, most of the cases coming through the stratigraphy.

 But of course, if you have better ways of describing the anatomy, you should to use that is the case of animals.

 Most of the cases are structural.

 The so-called connectome, the structure are in terms of the fibers, existing couplings.

 And then you put your favorite way of describing the local dynamics and you try then to explain the global dynamics, nothing else.

 This is called hybrid model it, of course.

 And you you have to define okay, but which aspect of the global dynamics.

 Traditionally, we were always obsessed with the function of connectivity, functional connectivity dynamics. And so a what I will use today is get out again when I say, for example, the inside out matrix.

 So the the, the label of of of asymmetry in the in the forward or backward version of the shifted functional connection.

 And I will try to explain that from the empirical data.

 So I will try to fit the model so that they explain the the so the top matrix is the biblical one and they will try to generate an explanation of that with the model for that.

 They will try in many parameters, namely I will associate to each existing fiber value and this is what we call effective connectivity.

 A And I added just because the term effective connectivity was used in many different contexts, I put generative affective connectivity because it's the one really coming from the from the model.

 So and, and we optimized that.

 And in this case we use our favorite the local linear, local nonlinear dynamic, which is the two Orlando oscillator and does some good reason for using that.

 An example first left of again model three HCB What you see in green is the box plot of all ties together now and not distinguishing them but calculated with the inside out.

 So they are at the top has the maximal hierarchy in the middle and blue light blue you see resting state.

 The difference that one is with the seven Tesla, the others with a three Tesla.

 So but very similar.

 And what is important that is they are a below the hierarchical organization of cognitive tasks as we found before with with Tenet, but now with the inside out.

 So that is is good that meaning that the methodology on the same data is achieved in exactly the same.

 But the most interesting case is the yellow box plot.

 The yellow plot is a condition where they are watching a movie which is very interesting and use a lot nowadays in neuroscience.

 And what we see is something that is unexpected.

 The theoretical organization goes down that caused me a couple of months in my life because I was absolutely convinced that is an error and I was doing the MATLAB and so on, say, come on, I am getting older.

 I mean, I'm not used to do I calculated within it.

 Also down below, I calculated with the dataset from, from Lausanne, from Switzerland, with young people watching movie all the time.

 So I started slowly to be convinced is really smaller.

 And then I got the interpretation.

 I mean, well, the first interpretation is, finally demonstrate why movies are so relaxing is because they are disconnecting the brain.

 But actually the truest nature.

 Of course, movies are relaxing, I think is the word resting state is absolutely misleading.

 Of course, during resting state we are doing much more things that we that that would that we are executing in movie watching.

 And it's confusing the word resting I mean that implicitly.

 Okay.

 So we have very very clear quantitative results.

 So we create a model in this case is one model for each single participant in each single condition.

 So one model for participant one watch in a movie or resting state.

 And then we take the GIC.

 So the generative affective connectivity and we try to prove, okay, this generative effect, the connectivity, the generators of the Iraqi, is the mechanistic explanation of the generation of the, you know, should be useful.

 How I showed that while the easiest way is let's try to classify so we use standard classifiers in this case support vector machine, for example, we try to classify the conditions if one particular person is watching a movie or is in rest.

 And as you see on the right, I mean, or just get the feeling by the box plot, which are very high, the level of accuracy of these classifiers is pretty high.

 So over 90% if you do with the functional connectivity.

 So forget thermodynamics, forget modeling, forget everything, just the absolutely stupid functional connectivity of your data.

 So just a correlation you doing also good job is to is a blue box blood a but not always.

 The conditions where you are practically just labor in particular is what you see right bottom.

 We classify there are two type of movies Hollywood movies and creative common movies, Creative Commons movies, YouTubers, things like that.

 And Hollywood are the Hollywood movies are Mission Impossible, James Bond, as all the A F sees, is very bad in classifying that.

 But the Jeep.

 So the generators of the you don't go when decision is over 90%, you can distinguish between the different type of movies.

 I would love to repeat that between Hollywood and European movies.

 But then I don't have to do it.

 I am sure that that defense same technology model base of the inside, out of the you are going imitation for the same dataset that I was showing before.

 We then it the biomedical dataset, the coma cases and doing the same of course in the box lot the is not shown here.

 The difference is like internet in the level of unionization.

 But when you do the model and you try to classify the model in controls minimal consciousness or uncommunicative wakefulness in drones or deep, deep coma, then you have a very good level of of separation.

 And if you want of of diagnosing really the type of stage of coma that that you have in that condition, which is of course a hot topic in in in coma research.

 I just want to finish with some speculations about the future, how we can far that apply really this these ideas of thermodynamics and now in a much more sophisticated way to get extra information in just based on the same conceptual idea.

 I mean in equilibrium we have no knowledge of time ever seen, not reversibility and so on.

 And in non-equilibrium we have something different.

 We have to give quantization to A and that is a theorem which if you are not from physics, even if you are from physics, I forgot that theorem.

 I mean, to be honest with you, when I started to look that again, it's a fluctuation dissipation theorem.

 Scardoelli mean, when you say flotation of what this equation of what?

 Well, that was the first important contribution of Albert Einstein when he was in Ban at the patent office.

 Doing bureaucratic work seems to be that he was extremely wor and he started to look at the what we call nowadays the Brownian movement.

 So you put in a glass of water, little the particles, and then he was really just looking at how they fluctuating fluctuation.

 And he explained fact that what happens as he was really very bored and then say what do what should they do now?

 Okay, let's put some charge on the particle.

 And I put an electric field.

 So I produced this system.

 And what he found is fabulous and is the and this is the dissipation part.

 Namely, that's a if the system is in equilibrium, the fluctuation, but addict hundred percent, the dissipation that is magic is not magic, of course.

 But but it sounds like magic.

 I mean, I just look how they fluctuate when I not produce the system.

 And that is enough.

 I can predict Perfect.

 I don't need to measure.

 I can predict what would happen when you party of the system.

 Fantastic parenthesis.

 Let's go to neuroscience, one of my essays in energy Partners.

 But it's not because of that.

 I discuss all that they got into that.

 And with this Marcelo, my cimini he had is a medical doctor, had a fabulous intuition and a fabulous idea.

 He has no idea about Albert Einstein, has no idea about the this patient.

 And many years ago they have this great idea, we will produce the brain with DMEs, for example.

 That's the easiest way I will characterize the brain in a in a very cheap way, e.g.

 so really cheap.

 And he had this intuition.

 That's when a part of the brain and they have different underlying dynamics because it's associated with different brain states, sleep, anesthesia, coma, whatever.

 Then the effect of the perturbation is different and then say how a measure that I have no idea.

 I never studied the physics or I have.

 So I used a more simple thing, which is fantastic.

 And this is what he called the PCI.

 The perturbation Complexity index are used to compress the ability.

 So the sleep that you have on your laptop, you don't need to understand even celibacy for complexity.

 So I use impulsive complexity.

 So I think how much I can compress the signal that I evoke after the perturbation and they managed to publish very relevant papers, nature signs and in many other journals showing basically what is the nice, what is not gotten.

 I too, from one of the early papers that's under different conditions like resting wakefulness or sleep and non-REM sleep or REM sleep or coma.

 What you see at the bottom right again, deep coma, meaning out of consciousness or anesthesia is different here.

 You see, even without the PCI, I am not the blood in the box, blood of the PCI hip. Lots to that, of course.

 But you see here just by the cartoon that the signal is simple and therefore is more compressible, the more unconscious you are.

 A fantastic idea.

 Really fantastic idea. Fantastic intuition.

 And they, they and they, of course, develop it that in the context of not only of of biomedical application, but in the context of consciousness research.

 Now I try to link and of the parenthesis I try to link the fluctuation dissipates in theorem with this and with the more dynamics and this is how I will finish.

 The idea is very simple and I hope that you already came to the idea that works because the system is non-equilibrium.

 If it would be equilibrium, then a massive mini would have discovered that I don't learn anything because I could distinguish them of course, but I do not need the perturbation.

 I just look at the fluctuation level and that should be enough.

 But it's not enough. Why?

 Because all those brains are in Non-Equilibrium and therefore the fluctuation dissipation theorem does not hold.

 And that means that the degree of violation of the fluctuation dissipation theorem or the degree of hierarchy, realization of the degree of arrow time, or the degree of non-equilibrium is a measure of the brain of states.

 And instead now of characterizing this brain state with data of time, as we have done during the whole talk, in all the different styles, model, free model based, I would use perturbation, and that is a beautiful way of doing that.

 So if the brain would be in equilibrium, what your is and I say to you that the spontaneous state, the resting state, and this mathematically that I see it in a very simple way.

 I mean, I don't go into the details, but it's just correlations.

 It's like the vanishing connectivity and the dissipation which is basically what we call susceptibility is the effect of the.

 So it's a more sophisticated version of the level of they should be equal.

 And that is the fluctuation dissipation theorem.

 But that is not the case because the brain is in non-equilibrium.

 And therefore what we want to see is how much we are contradicting this equality.

 So we define the measure, which is basically the difference of the left and right hand side that normalize it.

 And this is the degree of deviation from the fluctuation dissipation theorem.

 And this is a measure, another measure, a clever measure, because it's used in something that is not the in in the in the pure fluctuations, you are adding the defect of the dissipation of the perturbation.

 And of course, one can do this with the model exactly the same as before.

 And you produced the model in Silico in all these different way and just measure different anticipation deviation.

 And what you see here, just look at the box below.

 I mean, the renderings is that this is just about the we did, but just in the box plot, you see two cases sleep against Awake human neuroimaging small set 18 participant at the date of hand was love very well known and the data that we use it all the time?

 HCB More or less thousand participants in all this condition repressing cognition.

 So and the degree of violation of the voltage anticipation theorem is a good quantitative measure of the degree of non nicotine, so that you are justifying why the PCI works and how you can define the.

 It sounds a little bit arrogant, but it's not that the right version of PCI because it's not intuitive.

 It's the one that came in from first principles, namely the fluctuation, dissipation, delay and in that direction that we are trying to go.

 Now, they say I think they go very rapidly on the conclusion their main objective was hierarchical organization, corporate level, using whatever you use e.g.

 me, G, I and I, even local field potentials animals, humans in all possible conditions.

 And the trick was extremely simple.

 It just looked out of time with different methodologies, with machine learning, with correlations.

 so basically that is the main they commit out of time.

 Non-Equilibrium And you don't go any stage on the word turbulence.

 Forget it comes around. I did talk of that.

 I'm not mentioning today, but this is strongly related with that.

 And we have done a model of that and we proved that the generators know the parameters of that model, have efficient interpretation of these what we call generative effect. Connectivity is informative.

 Is that then something about the origin of this hierarchical organization?

 And nowadays we are trying to develop a sophisticated, sophisticated technique in the framework of the fluctuation dissipation data.

 So I was doing all this with my good friend from Oxford, Morten Kringle.

 Why we develop it, all these ideas, basically, do we think of it?

 We were also born, not invented.

 This at home.

 Was that really a pleasure to work with him online on all these things?

 So thank you very much for your attention.

 But wow, that's a lot of food for thought, Gustavo, and that I have a lot of questions that I want.

 One point.

 So you derive two indices, one through the reversibility component and then another one when you compare functional conductivity with this more elegant dissipation equation.

 And what I found interesting is the ranking of the cognitive states from the Human Connection Project is the same with the both techniques, isn't it?

 It's roughly the same. Yeah.

 So you said the extremes are the same social and emotional, such as is the maximum, the emotion is the minimum them in the squid.

 Interesting to me.

 So there's something tapping into something that you use.

 It's yes, I mean the the actual one can one can look really when can look the renderings of the local version of the arrow of time and then you see that that's really there much more asymmetry in a complex task like the social task of the HTP than in the emotional task, which is basically just visual perception.

 I mean, they call emotion because they distinguish as angry nothing but the that there is an emotional component.

 You see really the typical suspects.

 I mean that there are asymmetries in the interactions but that there is a much more massive asymmetries, relational or organized orchestration of computation in complex task than in simple task, as I suspect.

 And one thing that I didn't quite understand is how you do the perturbation on the last part. Yeah, yeah.

 So we take the model, individual participant, individual condition, we take the model in Silico and then we produce the model.

 So each single note is perturb it in some way.

 I inject noise.

 All right, Inject noise beginning right over the air, the beauty that you can do some analytical tricks because the model is linear.

 I see it.

 And and, you know, at the end of it, you have an analytical version of the dissipation.

 So therefore it is very rapid to calculate.

 Okay, You know, I was thinking it had to be knock out a node or inject something.

 Okay.

 And just one of the thing I want one point and with a lot of MRI function imaging measures is a single patient data is always very noisy to to classify a single patient as schizophrenia.

 You have one event.

 But I just wondering whether your global measure should the global measure of non reversibility for example how well that would work on an individual.

 Yeah.

 Indirectly I saw some results.

 So the schizophrenia result bipolar ADHD or the coma results, there were single single patients. Yeah.

 The dots were single patients.

 Yeah.

 All, all the analysis, even model for your model based single patients and the degree of and using a global measure you can use.

 Well the that is is pretty informative so the effective connectivity is very good but just a global measure of non-equilibrium in some case is very good to distinguish it.

 It's not so good for classifying of course is to course to block it.

 But if you go into the details and for example, one example is the cheek, the affective connectivity done, done the label of classification article and really high and really high.

 So that so that even with the with poor data, because the quality of the data are really not to encompass, you know, the neuroimaging.

 I mean these are next to the factor and if that's the brain is pretty distorted, Yeah, but is, is another source of noise.

 I mean that each individual is a nightmare.

 But in spite of that that you saw I think was over 80% the classification.

 Yeah.

 And that's what I was saying that was doing very well.

 I mean, all of these diseases as well, they share the same genetics and say schizophrenia and bipolar disorder and ADHD.

 So It would be surprising that you differentiate well between them and have the same functional networks as well.

 So that's why it struck me that it does it does well as a technique.

 Yeah, we didn't try to classify it, to be honest with you.

 The the UCLA, but we classify other psychiatric diseases or even smaller, for example, the depression that I said from robot Robin Carhart-Harris, the one that we actually use psychedelics.

 And so it's a beautiful thing because he uses citalopram that the traditional they said I drug for depression or the psychedelics and then you have both them and then you can see the effects of loss and it says model that I said because they have more or less 20 patients bad condition and that works very good.

 I mean, the classifier of we do classify it's not Jake is a little bit different.

 Image is another characterization of the hierarchy but not global.

 It's much more detailed is what we call trophic characterization and that classify it relatively well even can predict the response to the drug, which is it for me is the Holy Grail.

 Yeah. And so that's going to be looked for.

 So a lot of these things we don't have that much clinical applications for animal research despite the huge amount of MRI research that we had just squeakers for this one issue that is not so reliable on an individual patient basis, except okay, I have more questions, but I'm being greedy.

 So someone else went or something.

 Yeah.

 So thank you for this is an interesting topic and I would like to know your opinion about the the causality for all these type of molars.

 At the beginning you mentioned causality, but it seems to be not enough for several situations.

 So I don't know if you are familiar with the type of alternatives for so on causality for measured in the study.

 Yeah. Yeah.

 I don't know. And in fact I'm in the second version of the model for the inside out.

 It's a cheap version of the Granger Causality A and, and that was enough but things because we are not so much interested in the absolute value I mean perhaps this body but in characterizing the degree of interactions, if you are really interested on the values but is enough to assess the degree of asymmetry.

 So my my, my, my belief or my intuition nowadays is that thanks to this thermodynamic group, we are allowed to measure causality in a bad way.

 And that's would be enough, of course, if I if I would be able to find something better than the Granger causality or transfer entropy, I would use that.

 And if I have enough data, I would use that and I would then do the thermodynamic trick.

 I just would construct the graph and analyze in the traditional way.

 But I was not able to find that way for and robust enough for applying this to to individual patients.

 We have a really very bad quality of data, and this is that in that the example you gave of this reversibility with the Russian name at the beginning, what was the name of the reversibility of movies, the reversibility of movies in time?

 Are the the movie, you mean? Yeah.

 Christopher Nolan No, no, no.

 That, that someone that had shown that it was reversible, but over a small timescale just since, just introducing, you know.

 So what sort of timescales are we thinking of and what, when when they were, when they come up with this idea that was and it's been actually it's artificial timescales, I mean but the if, if you would apply these are exactly the same type of spin model that we use in zero image.

 And by the way from time to time.

 So you can apply these if you want, even at the millisecond scale if you want.

 Yeah.

 Yeah, I thought so.

 But I think what you are trying to, to formulate is a good question that I don't know the answer, but they will work against me and is actually is a thin is a challenging question if the level of non-equilibrium or if the level of communication changed with the timescale that I think I believe that of course I don't know.

 We never highlighted one can do that.

 We have not done that.

 For example, in the manga or with imaging.

 And then you really on purpose, you, you just concentrate on different bands or different filters and then you have the answer.

 Yeah, that is we have never done that.

 It's a good question about that. Yeah, but there's also a question.

 So we in electrophysiology we use Granger calls on a team and, and sometimes review.

 I sometimes happen to say have you, have you, have you flip the data around to make sure that there is no Granger causality.

 that should be a good answer goes I did but the thing so yeah the because basically showing that information flows in one direction so the history of area predicts the history of area B better than the history B itself on its own.

 Yeah I think I got that right.

 Anyway so just sort of regressive things and, and there should be and that's what I was going I was wondering okay so maybe, you know, you can be lucky and if you're doing with higher frequencies, you get one result.

 Then if you're dealing with low frequencies, when you flip the.

 Absolutely.

 Actually we have that of one of our reviewers also was asking, well, you are telling that thing that in a to cartoonish or it depends on causality similar to the cat you showed that and they have shown that I mean I took the ACP I calculated I had operated like an engine goes I did I calculated the degree of asymmetry of the Granger I was head of the classical Granger causality A and that correlates perfectly with the inside out.

 All right. Okay. Well, that's good to know.

 Yeah, it's a I was annoyed because it was extra work, but then we can refer to it and to review of comments, please see Decker's response on this reversibility of Granger causality.

 Yeah, there's another point here and that would then work in your machine.

 But and there's a lot of debate as to how much the structure, the function of connectivity in the brain matches the structural connectivity.

 So there is some gross component that and you know, the areas that have bigger connections will be fluctuation, the blood oxygen level up and down a bit more.

 But you go one step further with your modeling.

 So you're absolutely you use this the generative effect of connectivity defined by the structural connectivity.

 Yeah.

 So how how, how do you do that?

 So precisely.

 So it is every single node of your functional map have a have or have not got to a connection through fibers?

 No, no.

 The connections are only the existing connections.

 So therefore I take the the DTI with all the problems that I have, but I take these as the ground through for me.

 Okay, this is my template.

 The good news, because in in the cases where I used the defective connectivity, I only take the DTI as a mask, so I will only update the connection that exists.

 But the strain of that connection does not need to correlate with the number of fibers.

 Okay? And that is an advantage.

 And that therefore I give the name effective because I don't care about the anatomy.

 Axes are not axes, but I decide this strength based on the data, of course, but based on the functional data.

 And they get much more than the If you try to explain functional connectivity with the structure of connectivity, use, really interesting say you get it'll depend on the Barcelona and blah blah blah blah.

 But then for the standard facilitation, let's say the hundred, you get 0.3 a correlation, a model.

 So for example, this tool and the model in dressing state could go to 0.8 so that means that you are explaining much more than the structure, structural connectivity.

 And this is also known because that is a very actually very simple manipulation was done by the defend with monkeys and by and so Douglas look I love with humans one with Anastasia the other with the with the sleep if you compare just the correlation that I mentioned before see with FC you said 2.3, but if you check that in a sleep or in anesthesia goes up.

 All right.

 So I don't remember the number, I would say instead of point four, but significantly, both at and that is also entity and that the means of course the condition it's not I always say it's beautiful sentence from from Aristotle's that was translated Latin by queen and quickly through a GP to that models which is the GP do which means the container shape the content and say yes shape but is not determine if they can't.

 Right.

 So the see is really shaping FC but is not defining.

 So the dynamic is defined in the FC and therefore is not astonishing that neither sleep and and in non the sleep you have different values for me as demonstration that you see dynamics matters.

 Yeah.

 So the more the more active your brain is the less it's the less the less it, the more independent Are you from this.

 That is also a very naive view of that as a come on the anatomy is pretty fixed.

 Okay.

 We know that's changed a little bit, but but it's pretty fixed.

 And we are doing in every second of our life totally different things.

 So how how can be how could we be so flexible with the fixed, boring anatomy?

 Okay, well, if they.

 No more questions, just think, Gustavo.

 Once again, it is fascinating. Thank you.

 Pleasure. Thank you.