WEBVTT 00:00:03.537 --> 00:00:10.497 This is the Convergent Science Network podcast. Leading researchers in the domain 00:00:10.497 --> 00:00:16.737 of neuroscience, brain theory and technology are interviewed by Paul Verschure and Tony Prescott. 00:00:19.637 --> 00:00:23.817 So this is Paul Verschure with the Convergent Science Network podcast. 00:00:23.957 --> 00:00:28.677 Here we are at BCBT 2018 together with my colleague Tony Prescott. 00:00:28.797 --> 00:00:30.737 Hi, Tony. Hello. Good to see you. 00:00:31.397 --> 00:00:40.377 And we're here with Luis Fuentemila, who is a great researcher of memory. Right. 00:00:41.817 --> 00:00:43.677 So what's your definition of memory? 00:00:45.157 --> 00:00:49.037 I think memory, because it's kind of a complex question. 00:00:49.757 --> 00:00:57.517 But definitely the way I would perceive memory is the only one of the only functions 00:00:57.517 --> 00:01:01.457 that allows us to bring some persistence over time. 00:01:01.697 --> 00:01:08.957 So basically it links moment to moment into a continuity. That's the way I see it these days. 00:01:09.217 --> 00:01:14.937 Okay. But now the title of your talk was To Shape the Unfalling Experience into a Memory Code. 00:01:15.737 --> 00:01:21.177 So the idea of the title was a bit to kind of like point out. 00:01:23.217 --> 00:01:29.637 That at the end of the world, or at the end of the day, The way we create a 00:01:29.637 --> 00:01:34.397 world is based on the way we make representations of our inputs. 00:01:34.997 --> 00:01:38.017 And that transformation, which is like in essence one of the biggest problems 00:01:38.017 --> 00:01:43.077 nowadays to face, somehow should be shaped by a representational system and 00:01:43.077 --> 00:01:49.337 some mechanisms that should be framed into the memory, into the memory function, let's say. 00:01:49.737 --> 00:01:53.617 So if you put this argument into the extreme, I would say that somehow, 00:01:53.877 --> 00:01:56.517 right, the way you perceive or the way you see the world or interact with the 00:01:56.517 --> 00:01:58.997 world, it should be definitely shaped by a memory system. 00:01:59.437 --> 00:02:01.517 So it's kind of like a loop thing. 00:02:03.208 --> 00:02:09.948 So now, to start your talk, you also emphasize this whole issue that also a 00:02:09.948 --> 00:02:14.208 memory system is also actually about forgetting, right? It's about selectivity. 00:02:15.648 --> 00:02:20.528 So how do you see the selectivity of a memory system? What defines its selectivity? 00:02:21.148 --> 00:02:25.428 Well, I think this is one of the important puzzles that we are facing in the 00:02:25.428 --> 00:02:30.768 memory, because assuming that you're forgetting, it's quite logic, 00:02:30.908 --> 00:02:33.748 because we experience that, I think everybody would accept this. 00:02:34.388 --> 00:02:38.848 But on the other hand, we are failing to this fallacy of accepting that something 00:02:38.848 --> 00:02:41.988 doesn't exist because we cannot observe it. 00:02:42.528 --> 00:02:46.228 So at the end, like from a clinical point of view, it makes sense to assume 00:02:46.228 --> 00:02:51.808 that anytime you cannot recall things, it's because they are forgotten and that might have a function. 00:02:52.088 --> 00:02:56.208 But on the other hand, there's this sort of like a fallacy of assuming that 00:02:56.208 --> 00:03:00.488 this doesn't exist. And I think the more you, so I'm not going into the response 00:03:00.488 --> 00:03:03.308 directly to what you answered, to what your question was. 00:03:03.788 --> 00:03:06.008 But I think this is one of the fundamental problems is this, 00:03:06.068 --> 00:03:07.388 like, what do we really forget? 00:03:07.488 --> 00:03:10.768 And if we forget anything, because at the end, you can assume that at the end 00:03:10.768 --> 00:03:14.928 you create mental schemas, or you can transform into mental configurations that 00:03:14.928 --> 00:03:17.628 might also impact into the future. So that's on the one hand, 00:03:17.668 --> 00:03:18.208 the problem of forgetting. 00:03:18.944 --> 00:03:23.944 But on the other hand, I think that from like a mechanistic point of view, 00:03:24.084 --> 00:03:28.004 it makes total sense that somehow most of our information should kind of like 00:03:28.004 --> 00:03:31.264 disappear from our, to be able to come directly available. 00:03:32.284 --> 00:03:35.384 And then forgetting should be useful for this. So it has to be some sort of 00:03:35.384 --> 00:03:40.704 like a pruning process, at least from the availability point of view. 00:03:41.084 --> 00:03:46.064 And it kind of raises the question, what is memory for? I mean, why do we have memory? 00:03:46.704 --> 00:03:51.964 What's, I mean, what is the practical use for an animal to have memory or for a human? 00:03:52.604 --> 00:03:56.664 Well, I guess the one, maybe the way I see it now, at least with the research 00:03:56.664 --> 00:03:59.704 I'm currently doing is because it helps you learning more. 00:04:00.084 --> 00:04:05.084 It helps you interact like effectively and, uh, and, um, I wouldn't say take 00:04:05.084 --> 00:04:08.624 decisions because that enters into the decision making, the decision making world. 00:04:08.784 --> 00:04:13.464 Um, and I'm not like spotting into this, but, uh, but somehow the idea that 00:04:13.464 --> 00:04:17.824 memory, uh, basically shapes your experience because you assume that that should 00:04:17.824 --> 00:04:19.524 be useful in the future for something. 00:04:20.224 --> 00:04:23.584 So you make, right? It's kind of like, I guess it's like in the middle of a 00:04:23.584 --> 00:04:25.824 desperdictive coding scheme as well, somehow. 00:04:26.184 --> 00:04:30.724 I think there's a sort of contrast with our everyday idea of what memory is 00:04:30.724 --> 00:04:35.424 for, which is for retrieving things that have happened in the past and reliving them and so on. 00:04:35.584 --> 00:04:40.404 Whereas, in fact, maybe why the systems have evolved as they are is to help 00:04:40.404 --> 00:04:42.344 us in the here and now to make better choices. 00:04:43.624 --> 00:04:47.224 Yeah, exactly. Well, I guess that, yeah, that's kind of like an important question. 00:04:47.604 --> 00:04:52.524 So the extent to which memory just can be simplified, if you wish, or kind of like a, 00:04:54.564 --> 00:05:00.664 chunk into a kind of a function to take decisions or not, or whether this brings 00:05:00.664 --> 00:05:03.364 you out something else in the higher level functions like the self or, 00:05:03.384 --> 00:05:07.264 you know, like this sort of like a perseverance of your identity, if you wish. 00:05:07.444 --> 00:05:12.124 I think there's a mystery. three, I'm as a psychologist, I am not that into 00:05:12.124 --> 00:05:16.484 the idea of like, uh, assuming that memory is just simple function or it can 00:05:16.484 --> 00:05:18.704 be simplified as a function to take decisions for the future. 00:05:18.824 --> 00:05:21.544 I would like to think that there's something, it brings up something, 00:05:21.644 --> 00:05:23.224 something else as well, like the cell for instance. 00:05:23.913 --> 00:05:26.433 Which I'm not that sure, maybe we can discuss this possibly, 00:05:26.493 --> 00:05:30.293 to what extent the self or the agency can be also simplified. 00:05:30.833 --> 00:05:37.793 It exists because it helps us to take decisions for the future, optimal decisions. 00:05:38.233 --> 00:05:40.253 How about mental time travel? 00:05:44.973 --> 00:05:51.393 Well, I mean, that's kind of like the Endel-Tulbing suggestion from the very 00:05:51.393 --> 00:05:54.273 beginning, isn't it? Like the essence of episodic memory and what basically 00:05:54.273 --> 00:05:55.993 distinguishes us from animals. 00:05:58.193 --> 00:06:06.453 I sympathize with the idea that memory, it's the only way that we can just loop 00:06:06.453 --> 00:06:11.153 around time in a virtual world, which is extremely relevant for many of our 00:06:11.153 --> 00:06:12.413 interactions with that. 00:06:13.093 --> 00:06:17.853 I guess the crucial point is whether we need to be conscious about this constantly, 00:06:18.093 --> 00:06:22.393 the autonomic function of memory, or let's say property. 00:06:29.113 --> 00:06:32.913 In my view, actually, I think that actually this mental time travel like property 00:06:32.913 --> 00:06:37.813 is essential, and I fully believe that it's like a critical point for, at least in humans. 00:06:39.913 --> 00:06:46.693 Yeah. So in your own research that you discussed, the starting point were these models of memory, 00:06:47.453 --> 00:06:52.473 that had a dual process perspective, where you would say, look, 00:06:52.593 --> 00:06:57.613 we have acquisition of memory, two phases, and there's retention and expression of memory. 00:06:58.313 --> 00:07:03.773 Acquisition is the more hippocampal process, and long-term retention and also 00:07:03.773 --> 00:07:06.193 expression is a more cortical process. 00:07:06.553 --> 00:07:12.473 So why does dual process seem most relevant as a starting point for your research? 00:07:15.953 --> 00:07:20.753 Well, so basically this kind of like dual system model, it's like an ancient 00:07:20.753 --> 00:07:27.473 thing and it's a kind of traditional view from the 90s and I still think it persists heavily. 00:07:27.593 --> 00:07:31.073 And I would say that actually that's something that people like is taking it 00:07:31.073 --> 00:07:34.373 more and more and more serious or basically because they're proving why it seems 00:07:34.373 --> 00:07:41.493 to be a valuable point of view to address many of the problems that are needed to understand for, 00:07:41.653 --> 00:07:44.853 basically to understand how the memory is supported by the brain. 00:07:45.553 --> 00:07:51.873 So I like the idea of this dualism because memory should be a function that 00:07:51.873 --> 00:07:52.753 should deal with the present. 00:07:53.747 --> 00:07:57.767 Um you cannot just think about like i don't really like the idea of like thinking 00:07:57.767 --> 00:08:01.687 about memory as a an encoding process and a retrieval process as if you were 00:08:01.687 --> 00:08:05.927 these two were completely separate um because it's completely pointless in their 00:08:05.927 --> 00:08:07.467 life activity we're not just you know like. 00:08:08.707 --> 00:08:11.847 Basically canceling out everything from the surrounding and just retrieving properly 00:08:11.847 --> 00:08:14.607 or just incorporating information at least 00:08:14.607 --> 00:08:17.487 when we are grown because right the memory is always there 00:08:17.487 --> 00:08:20.387 and representations are there so at the moment that you are trying to 00:08:20.387 --> 00:08:23.047 simulate this is reality then you need a 00:08:23.047 --> 00:08:26.307 system that deals with this and uh this dualism 00:08:26.307 --> 00:08:29.087 seems to be at least from kind of 00:08:29.087 --> 00:08:33.147 like a right the very the principle or the kind of like the essence of the mechanism 00:08:33.147 --> 00:08:37.747 seems to be quite strict to understand it that's why i feel comfortable with 00:08:37.747 --> 00:08:41.527 that and i think that's why people now that basically are trying to bring uh 00:08:41.527 --> 00:08:45.167 this um conceptualization of memory as something that might deal with the present 00:08:45.167 --> 00:08:49.027 is comfortable with as well so the moment you bring like computational models, 00:08:49.327 --> 00:08:51.227 like you start talking about states, 00:08:53.607 --> 00:08:58.347 how states are basically relevant for decision-making, all these sort of questions 00:08:58.347 --> 00:08:59.647 that people are bringing up now. 00:09:00.087 --> 00:09:08.267 I think these models, that's why these models are so influential again and again. 00:09:08.627 --> 00:09:13.847 But now the experimental paradigm that you pursued focuses very much on this 00:09:13.847 --> 00:09:15.847 notion of reactivation, right? 00:09:15.947 --> 00:09:19.167 So in some sense, We have a simplified model so far in our minds, 00:09:19.227 --> 00:09:21.567 which is we have acquisition, retention, expression. 00:09:22.287 --> 00:09:27.367 But now you bring in additional components in that process, which is reactivation. 00:09:27.907 --> 00:09:34.747 So why do you think reactivation gives you the lever in understanding how memory functions? 00:09:35.747 --> 00:09:40.767 I think reactivation is based on the frame of this event these days. 00:09:41.267 --> 00:09:46.167 I'd say that it's a direct parallel of this sort of like a metaphor of, 00:09:46.187 --> 00:09:47.407 let's say, of a living machine. 00:09:48.087 --> 00:09:52.247 So it's one of the instances from a memory point of view and from this idea 00:09:52.247 --> 00:09:55.787 that there's these two systems that basically interact heavily at the moment. 00:09:56.007 --> 00:09:59.207 On the one hand, there should be an interaction, and this interaction should 00:09:59.207 --> 00:10:01.567 be based on a transformation of the inputs. 00:10:01.927 --> 00:10:04.967 Should be kind of like, right? I mean, it should be something, 00:10:05.207 --> 00:10:11.247 it should be, the two of them should talk into a language that should be memory-based, let's say. 00:10:11.447 --> 00:10:16.267 And reactivation helps us, or it helped me at least, to basically formulate 00:10:16.267 --> 00:10:20.907 a mechanistic hypothesis that fit quite well with this possibility. 00:10:21.907 --> 00:10:27.927 It's one of these mechanisms that really allowed me to think about how we instantly 00:10:27.927 --> 00:10:33.007 transform, rapidly transform any experience into a kind of like, let's say, memory. 00:10:33.564 --> 00:10:36.104 Memory buffer, very initially, like walking memory, if you wish, 00:10:36.284 --> 00:10:39.284 but also at long term, right, after sleep consolidation. 00:10:39.504 --> 00:10:44.084 That's why I kind of like to frame this mechanism as a kind of a critical ingredient 00:10:44.084 --> 00:10:47.804 for initial transformation of not just perceptual system itself, 00:10:47.964 --> 00:10:51.184 but for what we call, let's say, experience memory transformation, 00:10:51.304 --> 00:10:52.884 if you wish, something slightly more complex. 00:10:53.184 --> 00:10:56.304 And this replay can incorporate anything. I mean, at the moment that you create 00:10:56.304 --> 00:11:00.564 a parallel world, even if it's instant, rapid, and interacts with the rest of 00:11:00.564 --> 00:11:06.304 the brain, let's say, Then at that point, you can incorporate as many things as you wish, like goals, 00:11:06.564 --> 00:11:11.864 right, value, whatever you wish in there. And I think it's relevant because it's quite effective. 00:11:12.464 --> 00:11:18.624 I mean, would you say that replay was the mechanism for consolidation or are there other ways? 00:11:19.044 --> 00:11:24.744 Well, I guess it's not the only one. I wouldn't believe that that's the only one. 00:11:25.184 --> 00:11:30.844 It's one that fits well with the systems-level consolidation model. 00:11:31.464 --> 00:11:35.104 But for instance, there's another one, which is quite intriguing, 00:11:35.324 --> 00:11:40.884 in my view, from Giulio Tononi, like the synaptic myostasis hypothesis, 00:11:41.284 --> 00:11:46.164 which basically states that the brain has this sort of like a, 00:11:46.164 --> 00:11:50.304 or has this sort of like a threshold state that basically is lowered during 00:11:50.304 --> 00:11:52.824 the night, during REM session, that's what they claim. 00:11:53.044 --> 00:11:57.824 And at that time, anything that has not like the strength in connection that 00:11:57.824 --> 00:12:02.784 basically surpass this threshold dies and forgetting is based on that. 00:12:02.884 --> 00:12:05.024 So at the end, I think it's a combination of many mechanisms, 00:12:05.124 --> 00:12:10.964 but replay seems to be critical at least, you know, like to fit into this sort 00:12:10.964 --> 00:12:14.524 of like holistic model of how this hippocampus with the brain interacts. 00:12:15.024 --> 00:12:21.124 I think in sort of associative models of memory, you need something like replay 00:12:21.124 --> 00:12:25.724 because one-shot learning is not practical in most of those models, 00:12:25.804 --> 00:12:27.524 you know, neural network models, for example. 00:12:28.244 --> 00:12:32.224 Don't learn well off one example they have to repeat the 00:12:32.224 --> 00:12:35.524 example and they generally have to interleave it with lots of other examples to 00:12:35.524 --> 00:12:38.684 maintain balance so uh and 00:12:38.684 --> 00:12:43.024 this brings you to this uh opinion that you that you said that you have to have 00:12:43.024 --> 00:12:47.604 this fast learning one-shot learning system and then that trains the slow learning 00:12:47.604 --> 00:12:52.384 systems so you see a big difference there between the the way that the fast 00:12:52.384 --> 00:12:58.324 learning system will operate uh and how consolidation is happening elsewhere in the brain. 00:12:59.004 --> 00:13:02.464 Exactly. I mean, it's like simulating the rehearsal phenomena. 00:13:02.804 --> 00:13:06.584 So you train the other system to acquire this knowledge. 00:13:09.735 --> 00:13:13.755 I mean, still we assume that replay is a kind of like direct correspondence 00:13:13.755 --> 00:13:16.895 with input somehow. Yeah. Something that I'm not that sure yet. 00:13:17.255 --> 00:13:21.755 For instance, like we use this sort of like methodology with the classification approach, right? 00:13:21.775 --> 00:13:25.035 With the pattern classifiers, multivariate decoding algorithms, 00:13:25.235 --> 00:13:29.235 which is basically essentially you assume, right, that input or the elements 00:13:29.235 --> 00:13:30.355 or basically neural responses, 00:13:30.515 --> 00:13:39.055 neural patterns elicited during experience should be similar to those that basically 00:13:39.055 --> 00:13:42.555 are linked to the replay activity or to that. 00:13:42.775 --> 00:13:46.415 And that comes because we know the animal model with the face perception of the sequence. 00:13:46.675 --> 00:13:52.175 But I mean, right, the animal studies are slightly limited compared to human experience. 00:13:52.435 --> 00:13:57.175 And I think we are really like far away still to kind of like make this kind 00:13:57.175 --> 00:13:59.495 of bridge this gap, right? 00:13:59.495 --> 00:14:03.555 We take this assumption that sequenced pattern completion, all these sort of 00:14:03.555 --> 00:14:06.835 elements that involve replay that has been shown in animals, 00:14:07.035 --> 00:14:11.275 should correspond to our human experience, or we'll say our learning experience. 00:14:12.115 --> 00:14:15.475 And I'm not that sure yet about that. I'm not sure if anybody knows it yet. 00:14:15.855 --> 00:14:18.515 But we take this for granted, but I'm not sure. 00:14:18.935 --> 00:14:25.875 I guess that for the living machines, these sort of questions are quite relevant, 00:14:25.995 --> 00:14:31.115 isn't it? because for some of the projects you guys are working on these days. 00:14:31.335 --> 00:14:33.155 Well, I think the… This is the right point. 00:14:34.215 --> 00:14:39.275 Are we talking about… Absolutely. Consolidating, interaction, unidirection. Okay. 00:14:40.515 --> 00:14:46.195 All right. I mean, I think the process of consolidation can be thought of as 00:14:46.195 --> 00:14:51.015 multistage because your consolidation, you're still consolidating episodic memory. 00:14:51.655 --> 00:14:55.835 But at some point, where there are commonalities across episodic memories, 00:14:56.255 --> 00:14:59.075 you consolidate further into declarative memory 00:14:59.075 --> 00:15:02.555 and you lose the sort of maybe or 00:15:02.555 --> 00:15:05.915 you it's less important to have those memories 00:15:05.915 --> 00:15:09.815 specifically tagged to particular events that have happened so you learn something 00:15:09.815 --> 00:15:15.195 that's common across events so um mindy is that distinction useful to you or 00:15:15.195 --> 00:15:20.035 do you think everything you're studying is episodic no absolutely not actually 00:15:20.035 --> 00:15:23.415 that's one of the parts of for instance like we now we're running one of the experiment with uh. 00:15:24.481 --> 00:15:27.641 Uh, in a, let's say in a real world, which we ask people to, 00:15:27.661 --> 00:15:30.901 uh, kind of like these, uh, we use these wearable devices and we ask people 00:15:30.901 --> 00:15:32.381 to wear this camera for a long time. 00:15:32.441 --> 00:15:34.961 Because one of the puzzles is that when you bring people and you test, 00:15:35.141 --> 00:15:41.541 let's say episodic, uh, um, episodic, you ask them to recall episodic sequences of their own life. 00:15:42.161 --> 00:15:46.321 Um, most of the things are basically forgotten. They are, if you are into a 00:15:46.321 --> 00:15:48.881 routine they love, you go to the office, right? and you go home, 00:15:49.181 --> 00:15:54.481 you basically, things are so highly overlapping that you may say that most of 00:15:54.481 --> 00:15:55.581 your episodic memories are forgotten. 00:15:55.821 --> 00:15:58.781 So at the end of the day, especially with, so at the end of the day, 00:15:58.821 --> 00:16:02.641 when you test for the classic episodic memory recollection, let's say, 00:16:02.701 --> 00:16:05.381 this bivviness completely disappears. People are messing up. 00:16:05.541 --> 00:16:08.681 And we think that actually- Or they're using a schema to sort of fill out- And 00:16:08.681 --> 00:16:10.181 we need some, I think that actually. 00:16:11.761 --> 00:16:16.201 In the system, the way we replay information, the way we consolidate it, 00:16:16.201 --> 00:16:21.521 We need to put some more weight into this sort of like mental models that are 00:16:21.521 --> 00:16:23.021 slightly more relevant. 00:16:23.261 --> 00:16:27.361 Of course, episodic memory, I think in essence, it's extremely important and 00:16:27.361 --> 00:16:30.921 defines many aspects and it has a direct link with animals. 00:16:31.101 --> 00:16:36.701 But still in our day life, it might not have the impact that we may show. 00:16:36.701 --> 00:16:40.021 I'm not sure what this argument is, but it's a fact, right? 00:16:40.121 --> 00:16:42.601 I mean, we've been doing that. We ask people to wear these cameras. 00:16:42.721 --> 00:16:43.601 We do have like instances. 00:16:43.741 --> 00:16:46.841 We can even like record like short clips. We bring them back to home, 00:16:46.881 --> 00:16:48.741 to the lab. We ask for their memories. 00:16:49.741 --> 00:16:53.541 And they realize that it's their own life, but they cannot recollect the full experience. 00:16:53.681 --> 00:16:56.521 They cannot recollect the event. It's like, come on, guys, it's your own life. 00:16:56.641 --> 00:16:57.821 I mean, isn't it supposed to be? 00:16:58.001 --> 00:17:04.041 Well, it's another skill, isn't it, to actually to be able to convert memories into narrative. 00:17:04.761 --> 00:17:08.901 And I think the development, yeah, and the developmental literature shows that 00:17:08.901 --> 00:17:13.441 children right to school age are terrible at sort of telling stories about their 00:17:13.441 --> 00:17:18.241 lives, but they learn that. As adults, we get better at it. 00:17:18.321 --> 00:17:21.221 Some people are better at it. Well, it may illustrate your earlier point about 00:17:21.221 --> 00:17:22.261 memories also forgetting. 00:17:22.981 --> 00:17:25.481 Exactly. That's our story. Exactly. 00:17:27.101 --> 00:17:31.401 So you guys, for me, are moving really quick because you have an experimental 00:17:31.401 --> 00:17:38.241 paradigm where you test memory using reactivation and you do it using a conditioning paradigm, 00:17:38.421 --> 00:17:43.381 which actually is, again, bringing another possible confounding that we should look at. 00:17:43.381 --> 00:17:49.121 Because basically what you do is you have pairs of stimuli or sets of stimuli, 00:17:49.401 --> 00:17:56.461 one of which you condition your subject with a sound. 00:17:56.921 --> 00:18:03.881 And then you use the sound during sleep to, as you believe, reactivate that memory. 00:18:04.601 --> 00:18:10.241 And hopefully, through that, trigger whatever chain of items or memories it 00:18:10.241 --> 00:18:12.121 is related to. This is the core paradigm. 00:18:12.721 --> 00:18:17.861 And what you see with your healthy subjects is that you get about a 10% improvement 00:18:17.861 --> 00:18:22.361 in recollection of associated items. Mm-hmm. 00:18:23.032 --> 00:18:27.972 From that stimulus that you trigger with the sound as compared to stimulus that 00:18:27.972 --> 00:18:29.812 you don't stimulate, you don't reactivate. 00:18:29.892 --> 00:18:33.812 So now we know we could say, well, so you've shown that reactivation during 00:18:33.812 --> 00:18:39.512 sleep leads to a slightly improved recollection, 10%. Yeah, yeah. 00:18:40.132 --> 00:18:44.292 Imagine how it would be like to have more than 10%, right? We would have like the key of a... 00:18:46.772 --> 00:18:50.052 No, look. It's quite subtle there. 00:18:50.492 --> 00:18:56.572 Then, okay, so here we go. But now, curiously, does it matter when you play 00:18:56.572 --> 00:19:00.892 the sound, how long and what intensity, how often? What's the trick there? 00:19:01.912 --> 00:19:04.852 So one of the... No, that's a really important aspect, actually, 00:19:04.852 --> 00:19:08.112 of the design that I haven't mentioned in the presentation. I realized that. 00:19:10.992 --> 00:19:14.592 I was just halfway through of my life. 00:19:15.472 --> 00:19:19.312 I know what happened. No, seriously. I mean, I really wanted to present the last time. 00:19:19.672 --> 00:19:23.472 I don't know. I normally am not that bad in time. I mean, it's kind of like 00:19:23.472 --> 00:19:27.292 time-scaling thing. Somebody was interrupting. Oh, no, it was funny, absolutely. 00:19:28.392 --> 00:19:32.492 Ooh. Really? But, no. Right. 00:19:32.632 --> 00:19:37.852 So we are not sure about this kind of like intensity of the sounds. I'm not sure. 00:19:38.192 --> 00:19:42.512 I'm not aware of any kind of like study saying that the intensity has to be 00:19:42.512 --> 00:19:46.392 in a certain level so that, you know, like the trigger things during sleep. That's all. 00:19:47.092 --> 00:19:51.832 We definitely have to repeat the presentation every year, more than once. Repeat a few times. 00:19:52.532 --> 00:19:58.432 At least we repeat it like seven times each of the sounds per condition or per association. 00:19:58.912 --> 00:20:01.632 But we know from others that sometimes they have to repeat it even more. 00:20:02.132 --> 00:20:05.312 So that's an important aspect, you're right. But the critical point is something 00:20:05.312 --> 00:20:08.512 we control because we know it from many studies, 00:20:08.732 --> 00:20:15.192 is that at least it seems that to induce some sort of like beneficial effects 00:20:15.192 --> 00:20:19.472 of this sort of like artificial reactivation, let's say, design. 00:20:19.472 --> 00:20:29.492 Design, these sort of memories have to be specifically reactivated during slow-wave 00:20:29.492 --> 00:20:33.352 sleep, which is like one of the deep sleep, what they call, 00:20:33.532 --> 00:20:37.452 or one of these stages, like it's basically stage three and four, 00:20:37.632 --> 00:20:43.072 so it's basically taking place normally in the first half of the sleep. 00:20:43.372 --> 00:20:46.732 So sleep has this small architecture, kind of like with five stages, 00:20:46.972 --> 00:20:54.372 but there's two big sleep stages that are basically defined by brain activity, 00:20:54.832 --> 00:20:56.212 REM sleep, slow-wave sleep. 00:20:56.392 --> 00:21:01.772 Slow-wave sleep is the critical one because they claim there are few neurophysiological 00:21:01.772 --> 00:21:06.992 signals and responses and mechanisms that seem to be specially dedicated to 00:21:06.992 --> 00:21:10.892 promote this sort of interaction between hippocampus and cortical regions. 00:21:11.732 --> 00:21:15.392 Isn't that counterintuitive? Wouldn't you have expected that playing it during 00:21:15.392 --> 00:21:16.592 REM sleep would work better? 00:21:17.775 --> 00:21:23.015 Well, the point is that during REM sleep, it is believed that the cortical activity 00:21:23.015 --> 00:21:27.815 is highly active and is completely uncoupled to hippocampus. 00:21:27.855 --> 00:21:32.335 So then the model wouldn't work in that way because the model assumes that, 00:21:32.415 --> 00:21:38.835 right, this dual model system learning thing basically assumes that memories 00:21:38.835 --> 00:21:41.155 are somehow like initially stored in the hippocampus, 00:21:42.055 --> 00:21:43.475 probably together with our neocortex. 00:21:43.675 --> 00:21:47.695 And during this sort of like consolidation window, this optimal state, 00:21:47.995 --> 00:21:52.475 these two regions, or these two kind of like networks. 00:21:55.595 --> 00:22:00.555 Let's say, interact heavily, and that needs to be in a moment in which they do it. 00:22:00.715 --> 00:22:05.395 And probably, also, it's known that this is a moment where you are fully disconnected, 00:22:06.435 --> 00:22:09.415 from the environment, which is basically, you might focus a bit more, 00:22:09.495 --> 00:22:14.315 kind of like a tie-up, a tie-up, a bit, you know, like the mess you have in your brain, if you wish. 00:22:14.935 --> 00:22:19.075 So it's, uh, yeah, I mean, you say, because the REM sleep involves like a higher, 00:22:19.175 --> 00:22:23.715 let's say, more, um, cortical- So that would already point to the fact that as you like a system, 00:22:23.875 --> 00:22:30.195 memory in this case is a system property that exists in interaction with hippocampus and in cortex. 00:22:30.955 --> 00:22:34.715 Yes, exactly. Well, that's the idea of the model and that, I mean, 00:22:34.715 --> 00:22:38.115 there are many papers showing that that makes sense in animal studies with entorhinal 00:22:38.115 --> 00:22:42.015 cortex and with, uh, recently with, uh, with auditory areas using similar, 00:22:42.035 --> 00:22:43.695 similar sort of like designs. 00:22:44.315 --> 00:22:47.535 And in humans, with fMRI, there are a few papers as well. Also with kids, 00:22:47.695 --> 00:22:49.635 it was just recently published. Right. 00:22:50.235 --> 00:22:55.495 So some of the data you're pointing to are showing replay in hippocampus and 00:22:55.495 --> 00:22:57.455 replay subsequently in cortex. 00:22:58.035 --> 00:23:02.395 There was that in the sensory cortical area, the visual cortex, was it? 00:23:04.235 --> 00:23:11.015 So you interpret the replay in cortex as being about memory consolidation as 00:23:11.015 --> 00:23:15.695 well as the replay in hippocampus because the replaying cortex could just be 00:23:15.695 --> 00:23:16.955 the playing out of the memory. 00:23:17.615 --> 00:23:22.955 So the consolidation could happen in one or both places or in some other place 00:23:22.955 --> 00:23:23.715 where you're not measuring. 00:23:24.015 --> 00:23:28.815 No, you're right. I mean, this model, the model that basically these two systems 00:23:28.815 --> 00:23:33.535 are fully separated and basically start interacting in specific moments with 00:23:33.535 --> 00:23:36.515 geoconsolidated functions and then completely uncouple again. 00:23:36.715 --> 00:23:42.195 I think it's a bit extremistic, I would say, if you wish. I mean, 00:23:42.215 --> 00:23:46.455 it's not totally valid in the context of our interaction with the world. 00:23:48.167 --> 00:23:51.547 One of the things we know is that whenever you recall information, 00:23:51.847 --> 00:23:56.527 at least from my studies and some animal studies, whenever you're recalling vividly information, 00:23:57.147 --> 00:24:02.107 somehow like sensory related areas, maybe not like the primary areas, 00:24:02.227 --> 00:24:06.907 but sensory related areas linked to these memories tend to be active. 00:24:06.907 --> 00:24:09.727 Active um i mean in essence 00:24:09.727 --> 00:24:12.747 it would be like so the full idea is that uh right this kind 00:24:12.747 --> 00:24:15.467 of like vividness of your memories should be supported by some 00:24:15.467 --> 00:24:21.067 sort of like uh represent representative formats yeah so in that sense assuming 00:24:21.067 --> 00:24:26.827 that replay uh involves um representational stages of this information to help 00:24:26.827 --> 00:24:31.727 out transforming it it's not uh incongruent i mean when we're thinking Thinking 00:24:31.727 --> 00:24:33.567 about what's been consolidated, 00:24:33.947 --> 00:24:39.467 presumably we're imagining a very different signal in the hippocampus from the 00:24:39.467 --> 00:24:40.287 one that's in the cortex. 00:24:40.507 --> 00:24:46.087 The hippocampal one is going to be a very much compressed version of the memory 00:24:46.087 --> 00:24:48.667 and then it's going to be reconstructed in the, 00:24:48.807 --> 00:24:53.467 particularly the sensory cortex, it's going to be reconstructing something like 00:24:53.467 --> 00:24:57.207 the original stimulus or the effect that stimulus would have in cortex. things. 00:24:57.987 --> 00:25:01.827 So, I mean, for consolidation to be effective, 00:25:02.087 --> 00:25:05.087 I guess it does have to happen in multiple areas of the brain, 00:25:05.187 --> 00:25:14.427 or it does have to be the case that this system can reliably reconstruct based on a compressed trace. 00:25:15.587 --> 00:25:19.987 Do you think of it in those terms? Yeah, I think it's a really wonderful idea. 00:25:20.327 --> 00:25:23.567 You know, kind of like there's somebody mentioning that in this kind of like 00:25:23.567 --> 00:25:26.467 event, this idea that, um, uh, right. 00:25:26.667 --> 00:25:30.507 Like, um, uh, memory or memory representation, or let's say like a beautiful 00:25:30.507 --> 00:25:32.667 representation of the world. It's timeless. 00:25:33.367 --> 00:25:38.247 Right. And we give, we give some sort of like a structure, we give some sort 00:25:38.247 --> 00:25:44.167 of like a real sense of like a reality whenever you bring it to a singular sensory input, 00:25:44.247 --> 00:25:45.527 something, somebody just mentioned 00:25:45.527 --> 00:25:48.947 this possibility and I think I kind of like, uh, I think it was Josh, 00:25:49.107 --> 00:25:50.767 maybe something, maybe it was, you know. 00:25:52.290 --> 00:25:55.910 So how are these sort of like compression phenomena, I think it makes sense 00:25:55.910 --> 00:26:00.950 in this framework, so that the memory doesn't really care about time itself, 00:26:01.110 --> 00:26:03.170 for instance, it's just, you know, like a compact version. 00:26:03.790 --> 00:26:07.890 And then maybe at the moment that you interact with the system that is fully 00:26:07.890 --> 00:26:11.450 dedicated to interact with the world, then you have to give this other code. 00:26:11.610 --> 00:26:14.770 Is it what you are like? Well, the thing is, when you think it through, 00:26:14.830 --> 00:26:18.390 it gets rather complicated because you have of perceptual 00:26:18.390 --> 00:26:22.150 learning systems which are encoding signals 00:26:22.150 --> 00:26:24.930 sending them to the basis to the history the campus 00:26:24.930 --> 00:26:27.990 which is storing them in some way and then 00:26:27.990 --> 00:26:33.910 that's being retrieved and replayed and has to be decoded that decoding so that 00:26:33.910 --> 00:26:37.970 that encoding is being learned and can change over time so anything you've stored 00:26:37.970 --> 00:26:44.150 in the campus could become outdated and couldn't may not no longer match the 00:26:44.150 --> 00:26:46.830 decoding apparatus. So everything has to be kept in balance. 00:26:47.390 --> 00:26:53.170 And so all this sort of, in computer terms, offline learning that's going on 00:26:53.170 --> 00:26:58.430 during sleep may be about rebalancing all these different systems so that they're all in sync. 00:26:58.810 --> 00:27:03.570 Yeah. Yeah. Well, that's good to know. Louis, your second experiment you discussed 00:27:03.570 --> 00:27:05.050 also speaks to that, right? 00:27:05.110 --> 00:27:08.970 Because there you looked at the competition between memories where you had two 00:27:08.970 --> 00:27:13.250 kinds of associations, A and B, and B and C. 00:27:13.770 --> 00:27:23.050 You looked at how then triggering B would affect the reactivation-dependent recollection of A and C. 00:27:23.350 --> 00:27:27.590 But then what you looked at also is the time delay in that process. 00:27:27.870 --> 00:27:35.790 If you saw A, B three hours ago, and now I'm going to do a reactivation experiment. 00:27:36.230 --> 00:27:41.330 Does this time delay influence the memory? So is there a competition between memories? 00:27:42.250 --> 00:27:48.250 So I hope I summarized the experiment correctly because the point there was 00:27:48.250 --> 00:27:53.730 that you did see a competition between memories, as you call it. Right? That's correct. 00:27:54.030 --> 00:27:57.690 Correct. But do you see this as an active competition or passive? 00:27:57.990 --> 00:28:02.730 That means, let's say, since I saw something a while albacta might be as a decay 00:28:02.730 --> 00:28:05.970 of a trace, a non-specific decay. It's not regulated by anything. 00:28:06.610 --> 00:28:09.890 Alternatively, it might be an actively regulated competitive process. 00:28:10.170 --> 00:28:13.930 It really compares the activity levels and when it takes all, whatever. 00:28:14.170 --> 00:28:19.450 So how do you see this competitive process play out, given the data that you have? 00:28:20.570 --> 00:28:24.970 I think it's a really important point and that we couldn't disentangle. 00:28:26.510 --> 00:28:32.390 But we did observe something quite a realm that might speak about the possibility 00:28:32.390 --> 00:28:33.570 that there's something active. 00:28:35.050 --> 00:28:41.110 Although it contradicts a bit that the active phenomenon or active process could 00:28:41.110 --> 00:28:42.210 happen while you sleep, right? 00:28:43.685 --> 00:28:47.885 So in one of the points of the experiment, it was just a very fast slide at the very end. 00:28:48.065 --> 00:28:52.785 So we also studied how are the neural signals or neural oscillatory activity 00:28:52.785 --> 00:28:57.085 elicited by the sounds of the two conditions while people were sleeping. 00:28:57.285 --> 00:29:00.405 So people were completely unaware of the environment. They were fully, 00:29:00.405 --> 00:29:03.445 we could control for this, like slow-wave sleep were on track. 00:29:04.585 --> 00:29:08.845 And then we studied which were the neural responses when you gave the sound, 00:29:08.925 --> 00:29:09.765 when you presented the sound, 00:29:10.285 --> 00:29:14.465 and the condition which associated memories were boosted afterwards, 00:29:14.725 --> 00:29:18.105 and under the condition in which associated memories were forgotten, 00:29:18.245 --> 00:29:19.805 or kind of like forgetting was promoted. 00:29:20.385 --> 00:29:23.405 So it was exactly the same circumstance, exactly the same one. 00:29:23.545 --> 00:29:27.125 People were basically sleeping the same sort of sounds because they're counterbalanced, 00:29:27.345 --> 00:29:29.685 but the neural signals were completely different. 00:29:30.265 --> 00:29:35.225 So it seemed that actually the brain treated this input completely different 00:29:35.225 --> 00:29:37.105 as a function of what it has to be done. 00:29:37.225 --> 00:29:39.245 That was the interpretation we did because we had the Big Hero Effect. 00:29:39.245 --> 00:29:44.645 And that, of course, we could not attribute that as if it was an active or a 00:29:44.645 --> 00:29:46.305 passive. But it seems to me that. 00:29:47.187 --> 00:29:51.287 Somehow the brain entailed the memory reactivation phenomena, 00:29:51.527 --> 00:29:57.407 but when it detected that it competed somehow because of where we just claimed, 00:29:57.507 --> 00:30:02.247 we just hypothesized that it was because of the strength of this association 00:30:02.247 --> 00:30:06.867 could be potentially complicated to deal with in the future, 00:30:06.987 --> 00:30:09.427 then this activates also this other mechanism, 00:30:09.587 --> 00:30:11.707 this extra mechanism that we couldn't see in one condition. 00:30:11.707 --> 00:30:17.067 So like to, to, to, to kind of like summarize this point, we observed that under the same conditions, 00:30:17.307 --> 00:30:22.227 um, experimental conditions with the same sort of like stimuli and during sleep 00:30:22.227 --> 00:30:27.007 in one of the, the, the, the neural responses were slightly different and, 00:30:27.047 --> 00:30:29.867 uh, one component just appeared in one of the conditions, not the other one. 00:30:30.367 --> 00:30:35.847 So I think it seems to me that actually, um, these, uh, um, organizational principles 00:30:35.847 --> 00:30:40.207 are active, um, I'd say. And you would place them in hippocampus, 00:30:40.207 --> 00:30:41.627 in cortex, or in the redirection? 00:30:41.927 --> 00:30:45.467 And what's the mechanism behind that? How do you envision that? 00:30:45.767 --> 00:30:53.907 Well, I guess from a psychology point of view, the easiest way to see this is 00:30:53.907 --> 00:30:58.007 under the framework of this sort of induced forgetting phenomena, for instance. 00:30:58.387 --> 00:31:05.407 So there are these studies from Michael Anderson and others who basically state 00:31:05.407 --> 00:31:08.487 that whenever you inhibit yourself to retrieve information, 00:31:08.787 --> 00:31:14.407 right, actively, right, you stop your thinking about this, the consequence of 00:31:14.407 --> 00:31:19.767 this at the long term is that suddenly or eventually you might not recall this information anymore. 00:31:20.767 --> 00:31:24.467 This is like an effect that has been like framing to specific experimental design, 00:31:24.667 --> 00:31:28.807 and we know it depends on this, but somehow the possibility that this may happen 00:31:28.807 --> 00:31:34.467 also speaks of this idea that this active forgetting may take place in the brain 00:31:34.467 --> 00:31:39.347 and uh it helped us out understand uh this phenomenon at least in the context of this study, 00:31:39.827 --> 00:31:44.007 right um of course i mean i guess that like you know like um making the parallel 00:31:44.007 --> 00:31:45.767 of our experience where people were sleeping with. 00:31:46.719 --> 00:31:50.439 With induced forgetting, where people are putting an effort to forget this, 00:31:50.499 --> 00:31:54.499 it's a completely different story. But still, maybe the mechanism is similar. 00:31:54.859 --> 00:31:58.979 But that then also links to this idea you discussed in your third experiment 00:31:58.979 --> 00:32:00.979 on how to avoid memory clutter. 00:32:01.219 --> 00:32:05.919 What memory clutter would actually mean, that you actively also forget or suppress. 00:32:06.219 --> 00:32:09.979 Sure. I think actually that the way we understand or the way we see forgetting 00:32:09.979 --> 00:32:11.039 is extremely essential. 00:32:11.039 --> 00:32:15.519 To um so then the operation of forgetting allows you to understand, 00:32:16.199 --> 00:32:18.999 right that the memories are organized some way 00:32:18.999 --> 00:32:22.519 otherwise I mean thinking about like you know like the things that are associated 00:32:22.519 --> 00:32:27.139 considering that most of our experiences especially when we are like experts 00:32:27.139 --> 00:32:31.259 in learning and we acquire so much knowledge the links that we can make with 00:32:31.259 --> 00:32:36.359 things are so high and so at any point that we need some sort of like principles that govern this this, 00:32:37.219 --> 00:32:42.179 problem and forgetting it's extremely useful isn't it One of the principles 00:32:42.179 --> 00:32:46.539 that you pointed to was what you called the non-monotonic plasticity hypothesis. 00:32:47.119 --> 00:32:54.079 And that suggested that the sort of generalization that could happen occurred 00:32:54.079 --> 00:32:59.059 to similar stimuli, but then there was a dip. 00:33:00.159 --> 00:33:05.579 And so what occurred to me thinking about that was one of the challenges you 00:33:05.579 --> 00:33:09.059 have in episodic memory is to distinguish memories that are similar in some 00:33:09.059 --> 00:33:11.159 way, but that you need to keep separate. 00:33:12.179 --> 00:33:18.419 Sort of what we would call pattern separation in neural networks. 00:33:18.419 --> 00:33:24.299 And that this non-monotonic plasticity is maybe about keeping these memories 00:33:24.299 --> 00:33:30.639 separate that you need to remember what was different about coming to the lab 00:33:30.639 --> 00:33:32.879 today compared to coming to the lab yesterday, 00:33:32.979 --> 00:33:36.299 even though much of the context is the same. 00:33:37.999 --> 00:33:43.019 Is that a useful concept for you, thinking about pattern separation as a key 00:33:43.019 --> 00:33:50.199 thing here, and where also you might want to have active mechanisms for pushing patterns apart. 00:33:51.059 --> 00:33:56.179 Yes, I think this is like this kind of duality, kind of like the pattern separation, 00:33:56.399 --> 00:33:57.179 pattern completion phenomenon. 00:33:57.659 --> 00:34:01.979 It's an essential mechanism to make us understand many aspects, 00:34:02.139 --> 00:34:07.619 like how things that happen in the same context can still be represented separately. 00:34:08.379 --> 00:34:12.899 I mean, it's difficult because, as you say, it's very context-dependent. Yes, absolutely. 00:34:12.899 --> 00:34:16.299 So some things you want to see them as the same for uh depending on 00:34:16.299 --> 00:34:18.999 the question you might want to see it as the same effect there's no 00:34:18.999 --> 00:34:21.899 monotonic uh plasticity hypothesis that you mentioned 00:34:21.899 --> 00:34:27.259 from newman and norman to me seems to say more about active forgetting because 00:34:27.259 --> 00:34:32.839 you're speaking of a threshold plasticity if you fall below the threshold then 00:34:32.839 --> 00:34:37.019 then you you get the question right you really you dissolve those connections 00:34:37.019 --> 00:34:39.699 so this would not necessarily get you better separation, 00:34:40.219 --> 00:34:45.159 more pattern sharpening, removing the. 00:34:47.133 --> 00:34:52.353 The low frequency components of it, that you sharpen and you want it. 00:34:52.633 --> 00:34:57.853 The initial idea of this model, when they basically proposed it, 00:34:57.913 --> 00:35:03.453 was to try to explain the phenomena of active forgetting in the sense of their experiments. 00:35:04.933 --> 00:35:07.553 I think actually the pattern separation and pattern completion phenomena, 00:35:07.793 --> 00:35:10.653 which is a complex thing, and especially to be tested in humans, 00:35:10.793 --> 00:35:14.693 is still like many people that might complain about the possibility to quantify 00:35:14.693 --> 00:35:17.853 this in humans, even though there are wonderful experiments around the world 00:35:17.853 --> 00:35:19.473 on this topic with humans. 00:35:19.893 --> 00:35:24.333 But even though there's kind of like a concern, there's also something that 00:35:24.333 --> 00:35:29.213 should be incorporated in these models, which is like, because the essence here 00:35:29.213 --> 00:35:31.773 is what is the elements of experience, 00:35:31.933 --> 00:35:34.873 what is the contextual elements of experience, because the two are basically 00:35:34.873 --> 00:35:37.453 playing an essential role in pattern separation, pattern completion, 00:35:37.553 --> 00:35:38.633 the way we conceptualize this. 00:35:38.633 --> 00:35:45.413 But the way we understand context many, many times involves like our physical surroundings, right? 00:35:45.513 --> 00:35:48.953 Or you can even like jump it up to a mental schema, if you wish. 00:35:49.073 --> 00:35:50.853 Like, I don't know how the hierarchy can go up. 00:35:51.113 --> 00:35:56.413 But there's this sort of like idea that there's this kind of like temporal context as well, right? 00:35:56.573 --> 00:36:02.693 And in fact, it exists, this model I call like temporal context modeling from Aikohana. 00:36:02.693 --> 00:36:08.193 And I think this temporal context idea is also fundamental and has not been 00:36:08.193 --> 00:36:13.273 treated well it's been treated like kind of fairly but has not been at least 00:36:13.273 --> 00:36:17.393 to my understanding has not been like incorporated into pattern completion pattern 00:36:17.393 --> 00:36:19.353 separation phenomena the sense that, 00:36:20.181 --> 00:36:26.701 we can tease apart phenomena, things that happen in the same context in different days. 00:36:27.701 --> 00:36:31.401 That's one of the puzzles. But maybe the temporal context might explain quite 00:36:31.401 --> 00:36:35.561 easily because the time domain of these experiences are that simple. 00:36:35.801 --> 00:36:39.341 And the way we code for time, which is another important aspect, 00:36:39.701 --> 00:36:42.201 might help us understand it. 00:36:42.541 --> 00:36:47.481 Wait, just from my understanding, earlier you were saying the memory system 00:36:47.481 --> 00:36:49.301 you study is agnostic about time. 00:36:49.421 --> 00:36:52.281 It just time warps everything to make it really very compact. 00:36:52.481 --> 00:36:55.961 Well, that was one of the ideas that one of the speakers suggested. 00:36:56.481 --> 00:37:04.801 And I thought that was wonderful, but I already came up with some of the findings 00:37:04.801 --> 00:37:06.881 that people are finding these days. 00:37:09.021 --> 00:37:13.261 The existence of time cells, what they call them. So basically, 00:37:13.361 --> 00:37:20.981 cells in hippocampus, which basically seem to be coving like temporal gaps between intervals. 00:37:22.441 --> 00:37:28.521 I mean, I kind of like the idea that time should be represented differently 00:37:28.521 --> 00:37:32.761 in the brain and the memory system kind of like does not rely on the way we understand time. 00:37:33.281 --> 00:37:36.741 But still, right? It seems... If you want to do mental time jowl, 00:37:36.861 --> 00:37:38.701 you better do it faster than real time. 00:37:39.601 --> 00:37:45.901 It won't help you very much. This whole separation between the time warped and 00:37:45.901 --> 00:37:48.161 compressed to represent the more event and order. 00:37:49.159 --> 00:37:53.799 In which it is separately again, let's say, inflated time-wise to bring it back 00:37:53.799 --> 00:37:56.459 to behavioral time, okay, that would make sense. 00:37:56.639 --> 00:37:59.519 But I think as a memory system, you want to compress. You don't want to forget. 00:37:59.639 --> 00:38:00.499 You want to get real information. 00:38:01.219 --> 00:38:07.399 So maybe you don't want to have, like say, the real-time interval mixed in with the event sequence. 00:38:07.659 --> 00:38:12.459 Yeah. In fact, when you explore autobiographical memory, abilities of people 00:38:12.459 --> 00:38:17.879 in these experiments I've been in, and discussions with other experts or experts 00:38:17.879 --> 00:38:20.039 on autobiographical memory field, 00:38:20.239 --> 00:38:24.759 you just realize that the first thing that people forget is this kind of like 00:38:24.759 --> 00:38:26.299 temporal context information, right? 00:38:26.899 --> 00:38:30.719 One of the things, one of the properties of experience that seems to be disappearing 00:38:30.719 --> 00:38:35.699 quite rapidly compared to others, like salience, novelty, spatial context seems 00:38:35.699 --> 00:38:39.039 to be quite, whenever you can recall information, seems to be quite preserved. 00:38:39.279 --> 00:38:43.619 But the temporal information, which experience was before, when was exactly, was like, 00:38:43.719 --> 00:38:46.699 I mean, you have the sort of sense sense that it was not yesterday and it 00:38:46.699 --> 00:38:49.719 was whether it was yesterday it was one week ago but you 00:38:49.719 --> 00:38:52.439 are right you're missing uh it seems that it's something that 00:38:52.439 --> 00:38:57.219 disappears quite rapidly but that's uh right that's uh that's i like that and 00:38:57.219 --> 00:39:04.759 i agree with you without so too um so before we really move to the mag uh work 00:39:04.759 --> 00:39:10.659 that you did on on memory um so we started out with a dual process model, right? 00:39:10.819 --> 00:39:14.739 It's like an acquisition and retention model, the hippocampus cortex, 00:39:15.179 --> 00:39:17.439 right? It's not necessarily saying they work together. 00:39:17.599 --> 00:39:19.679 It's saying, first hippocampus, then the cortex. 00:39:20.399 --> 00:39:25.359 So, given the experiments we've looked at so far, what should be our conclusion 00:39:25.359 --> 00:39:29.359 to how cortex and hippocampus work together to realize memory? 00:39:33.819 --> 00:39:36.099 Let me try to get the point. 00:39:39.299 --> 00:39:47.079 So maybe one way to think about these two systems in the context of Visiband 00:39:47.079 --> 00:39:53.379 is that the hippocampus is essential, is the very first one to generate a virtual world. 00:39:54.679 --> 00:40:03.659 And without that, it would be completely impossible to kind of like, 00:40:03.859 --> 00:40:06.599 no, that's what I would think, actually. 00:40:07.199 --> 00:40:09.319 Where is the item information now? 00:40:10.619 --> 00:40:13.679 Which item? Sorry. Well, also your task. I have items. 00:40:14.099 --> 00:40:18.939 Oh, okay, okay. Yes. So we have an episode, and it's actually a number of items 00:40:18.939 --> 00:40:23.959 tied together, right? So you see that also as a hippocampal process? 00:40:24.919 --> 00:40:29.859 I think it's definitely in both, as if it was like trying to resolve the binding problem. 00:40:30.239 --> 00:40:33.679 Whether it's stored there or it requires the neocortex for that. 00:40:35.368 --> 00:40:41.488 Okay. So would you say they worked, cortex and hippocampus worked together from 00:40:41.488 --> 00:40:45.328 beginning till the end of the existence of this membrane? 00:40:45.668 --> 00:40:49.028 Yeah, yeah. Okay. So this whole idea of sort of traditional models, 00:40:49.168 --> 00:40:56.788 like hippocampus acquires and cortex retains, that you don't support now, given the data you have? 00:40:57.068 --> 00:40:59.848 I wouldn't say so, right? That's fine. 00:41:01.068 --> 00:41:04.668 I wouldn't say so at all, actually. I think it's a slightly more, 00:41:04.748 --> 00:41:08.448 the interaction between the two are more relevant that we can think about. 00:41:08.868 --> 00:41:13.248 And then you would put, what would be the relative contributions of the two? 00:41:14.068 --> 00:41:17.248 If you would have to give them a functional label, what would the hippocampus 00:41:17.248 --> 00:41:18.788 contribute, what does the cortex contribute? 00:41:19.288 --> 00:41:23.228 I think actually that representation itself, like the way we understand, 00:41:23.448 --> 00:41:26.288 similar to what Tony was mentioning. 00:41:26.508 --> 00:41:30.908 So the idea that the representation of like something that resembles truly related 00:41:30.908 --> 00:41:34.148 to the inputs we had must be stored somewhere in your codex. 00:41:34.208 --> 00:41:36.648 And the hippocampus should be free to do other sort of like things. 00:41:37.088 --> 00:41:41.068 Could work for instance, like as an index, as a pointer, something that allows 00:41:41.068 --> 00:41:46.368 to kind of like bind, distribute the information we need, because we need to 00:41:46.368 --> 00:41:49.528 recall for anything, we need to give agency, we need to give value, 00:41:49.608 --> 00:41:50.268 all these sort of things. 00:41:51.448 --> 00:42:00.408 It also may, we would, But the hippocampus might also be an optimal place to do comparisons, 00:42:00.768 --> 00:42:04.968 rapid comparison of what you expect priors with current inputs, 00:42:05.208 --> 00:42:10.268 and basically to drive signals to update models, for instance. 00:42:10.608 --> 00:42:16.528 So this sort of like computational issue should be driven by the hippocampus. 00:42:16.988 --> 00:42:20.568 That's what I think. While the representation itself, the way we understand 00:42:20.568 --> 00:42:25.028 representation in the way might might be better stored in the cortex. 00:42:25.648 --> 00:42:28.108 That's the way I understand that. Okay. So that's why at the end, 00:42:28.148 --> 00:42:31.208 we always see the two of them connected because they always work together. 00:42:33.108 --> 00:42:37.568 Okay. So now we have an idea of the system, right? And so now we can complexify, 00:42:37.608 --> 00:42:39.848 right? Because then you start to look at MEG. 00:42:40.988 --> 00:42:45.728 And then what you observed is that if you just look at the frequency distributions 00:42:45.728 --> 00:42:48.528 or the power spectrum of your MEG data. 00:42:50.408 --> 00:42:54.728 That better performance in these recall tests tests, was correlated with a much 00:42:54.728 --> 00:42:59.068 more pronounced and coordinated response in theta, let's say somewhere around 00:42:59.068 --> 00:43:04.908 8 Hz or so, as compared to an impoverished recall, 00:43:05.228 --> 00:43:10.208 where you saw a much more distributed distribution across the power spectrum 00:43:10.208 --> 00:43:13.868 of energy in the different frequencies. 00:43:14.328 --> 00:43:18.828 So is that a significant option? This is informative for us now, 00:43:18.908 --> 00:43:21.268 if we start to think about this hippocampal particle system. 00:43:21.268 --> 00:43:24.728 So now suddenly, Tata pops out in a very discreet time window, 00:43:24.788 --> 00:43:27.968 well, two seconds, I think, after someone's presentation. 00:43:28.728 --> 00:43:33.008 So what's that telling us now about this memory system? How is that working? 00:43:34.288 --> 00:43:39.348 So I think actually the relevance of this finding is that it basically expresses 00:43:39.348 --> 00:43:44.488 these, or it puts some evidence in humans, which was the point at that time. 00:43:44.988 --> 00:43:50.068 It put this evidence in humans by showing that there's some sort of like principle 00:43:50.068 --> 00:43:50.928 of organizational information. 00:43:51.268 --> 00:43:55.128 If you understand information representation as if it could be captured by a 00:43:55.128 --> 00:43:58.248 pattern, because if I, which is something that could, right, be endless discussed. 00:43:58.748 --> 00:44:02.468 But assuming that this is like something that we can argue, you. 00:44:04.968 --> 00:44:11.508 It basically provides a mechanistic model that basically confirms the idea that 00:44:11.508 --> 00:44:14.888 we need some sort of organization during maintenance. 00:44:15.228 --> 00:44:22.428 Otherwise, information wouldn't be preserved in a bound manner, if you wish. 00:44:22.868 --> 00:44:29.588 Okay, but does that bring you to this notion of an enlistment of a Tertigamma code from memory? 00:44:29.808 --> 00:44:32.668 That was the point from the very beginning. that was the point so the idea is 00:44:32.668 --> 00:44:37.048 like conceptually speaking we didn't advance so there's no conceptual advancement 00:44:37.048 --> 00:44:39.828 compared we just so that it was 00:44:39.828 --> 00:44:44.008 a much more methodological tool the force let's say because at that time, 00:44:44.708 --> 00:44:48.308 this model of this sort of like representation or this sort of like mechanism 00:44:48.308 --> 00:44:50.008 that are tightly related to representations 00:44:50.008 --> 00:44:53.968 were not couldn't be tested in humans at least as far as we knew, 00:44:55.468 --> 00:45:01.988 and I think the data was was quite nice but completely in line with the model, at least at that time. 00:45:05.139 --> 00:45:07.979 I think it's relevant because it also, like, breaches the gap with, 00:45:07.979 --> 00:45:12.019 like, mechanistic models that were kind of, like, shown in animals in many aspects. 00:45:12.239 --> 00:45:16.119 And it confers some relevance to the idea that this kind of, 00:45:16.119 --> 00:45:23.399 like, a face-coding mechanism that seems to be quite that relevant in many memory functions. Right. 00:45:25.739 --> 00:45:29.379 So now, for instance, like, so that paper came out, like, 2011, 00:45:29.379 --> 00:45:31.639 I think. It's kind of like an old one now these days. 00:45:32.159 --> 00:45:37.379 And I know from some of the work with it, afterwards, because at that time we 00:45:37.379 --> 00:45:41.439 were one of the first doing these sort of things, but now more people are working 00:45:41.439 --> 00:45:43.539 on this. We've discussed that. 00:45:43.699 --> 00:45:48.199 Maybe not linking it with theta, but somehow it seems that this sort of dual 00:45:48.199 --> 00:45:55.139 approach between oscillations and this sort of possibility to search for representation, 00:45:55.419 --> 00:45:57.019 which is another question, right? 00:45:57.159 --> 00:46:00.379 So this kind of tool, or this sort of combination of tools. 00:46:02.439 --> 00:46:06.959 So the universe looks like very consistent, right? 00:46:07.059 --> 00:46:13.339 So we have an idea about reconsolidation or reactivation, how we can use that for consolidation. 00:46:14.139 --> 00:46:19.819 We have an idea how hippocampus and cortex work together. We also see signatures 00:46:19.819 --> 00:46:24.139 of dynamics that look comfortably close to data gamma coding. 00:46:24.319 --> 00:46:26.239 So things look consistent, right, and manageable. 00:46:27.159 --> 00:46:30.759 But then you start doing these decoding experiments. with your MEG data, 00:46:30.839 --> 00:46:36.159 where you looked at the reoccurrence of frequency patterns in your MEG signal, right? 00:46:36.219 --> 00:46:39.719 So you were measuring your MEG over many leads, I don't know how many hundreds 00:46:39.719 --> 00:46:41.139 you had. No doubt. Plenty. 00:46:42.099 --> 00:46:46.159 Then from there you extracted, if you will, templates of frequency responses, 00:46:46.939 --> 00:46:52.459 and then you just used the classifier to say, well, if I'm acquiring the memory. 00:46:54.138 --> 00:47:02.018 Can I see any kind of echoing of that frequency pattern in a memory interval 00:47:02.018 --> 00:47:06.838 as I'm waiting to perform a recall test? This was the experiment you performed. 00:47:08.178 --> 00:47:12.658 And then what you showed to us was surprisingly complicated. 00:47:13.038 --> 00:47:18.238 And also surprisingly complicated because this is MEG, so you sit at the outside of the skull. 00:47:18.678 --> 00:47:22.498 So now we measure a very average signal across all brain structures, 00:47:22.838 --> 00:47:25.358 not only hippocampus anymore, right? The whole brain is important. 00:47:26.098 --> 00:47:31.778 So, on the one hand, it's amazing because you could really recover a pattern. 00:47:31.998 --> 00:47:35.638 The patterns were replayed in the waiting period, right? 00:47:36.398 --> 00:47:41.358 Before the recall test was applied. So yes, you saw the dynamic pattern that 00:47:41.358 --> 00:47:45.798 the brain formed to store the memory is retained in the waiting period. Okay, great. 00:47:46.938 --> 00:47:51.778 But on the On the other hand, we would also say, yeah, and that pattern is occurring everywhere. 00:47:52.298 --> 00:47:55.378 It's across the whole of Clifton because, as far as the data, 00:47:55.798 --> 00:48:00.918 it was not very restricted to, let's say, the temporal lobes, correct me if I'm wrong. 00:48:02.258 --> 00:48:07.218 So is that not now really some rocking the boat in a rather dramatic way? 00:48:07.518 --> 00:48:13.058 We were doing so well. We looked so neatly organized, and now we have this MEG 00:48:13.058 --> 00:48:18.638 classification result, which It seems to throw everything again in disarray. 00:48:19.018 --> 00:48:20.178 So what do we do with that? 00:48:21.758 --> 00:48:26.758 That's actually, at that time, that was one of the things that we discussed heavily in the group. 00:48:28.658 --> 00:48:31.838 So we assume that there's some representation. We assume that there should be 00:48:31.838 --> 00:48:33.298 some coding mechanisms for the representation. 00:48:34.138 --> 00:48:39.858 But other than that, we don't know anything. Or at least not that we can make 00:48:39.858 --> 00:48:42.098 any strong hypothesis about. 00:48:43.258 --> 00:48:46.778 One would say, for instance, It's like, well, because it's an offline representation, 00:48:47.058 --> 00:48:49.998 that should not involve sensory systems, for instance. 00:48:50.358 --> 00:48:53.078 But that's not the case. I mean, people have shown that in animals and humans. 00:48:53.458 --> 00:48:56.738 Right, okay, let's forget all that. So then let's think, actually, 00:48:56.818 --> 00:48:59.938 this information should be kind of represented in the higher visual areas, 00:49:00.158 --> 00:49:05.358 for instance, which it's true, but it might still affect other areas, 00:49:05.418 --> 00:49:07.918 associative areas, or frontal regions, for instance. 00:49:08.298 --> 00:49:12.558 So at that time, to be honest, we didn't know how to deal with this, 00:49:12.558 --> 00:49:14.318 and we still see it complicated. 00:49:14.418 --> 00:49:17.238 I wouldn't say we didn't resolve this. 00:49:17.338 --> 00:49:20.478 Actually, we kind of discussed this with one of the reviewers and we said, 00:49:20.558 --> 00:49:23.878 yeah, I mean, that's the way it is. That's the way we saw it. 00:49:24.018 --> 00:49:26.258 And we think that it's extremely complex. 00:49:27.338 --> 00:49:31.738 I mean, at least in that paper, we used at least more than 50 frequencies. 00:49:35.138 --> 00:49:38.738 We included more than 50 frequencies. From delta to gamma at all? 00:49:38.878 --> 00:49:42.718 No, we didn't include theta, up and that uh and values our coupling measure 00:49:42.718 --> 00:49:47.838 um and so we increased from like i think it was 14 hertz up to like 80 or something 00:49:47.838 --> 00:49:50.498 like these other like you know like the highest ones never um, 00:49:53.689 --> 00:49:57.449 But at the end of the day, unless you use some sort of clustering algorithm 00:49:57.449 --> 00:50:03.229 that allows you to slightly summarize into MEG in a statistical way, 00:50:03.409 --> 00:50:05.769 you always see so much noise in the data. 00:50:06.009 --> 00:50:09.769 We didn't implement any sort of clustering approach because we didn't know what to observe. 00:50:10.009 --> 00:50:14.209 And at that time, we didn't know how to deal with the decoding algorithm that well. 00:50:14.469 --> 00:50:18.129 I mean, we just jumped into using this sort of deep net thing or hidden layers, 00:50:18.369 --> 00:50:19.949 which was super complex. 00:50:20.469 --> 00:50:22.209 That was another problem of the project. 00:50:24.289 --> 00:50:30.129 Because we used the hidden layer algorithm, we couldn't be fully certain about 00:50:30.129 --> 00:50:34.589 which were the patterns that were basically accounting for the reactivation. 00:50:34.789 --> 00:50:37.889 So at the end, we kind of like inferred that because we show some of these patterns 00:50:37.889 --> 00:50:39.029 in the supernatant material. 00:50:39.369 --> 00:50:43.729 But we never wanted to make any argument because methodological reasons didn't 00:50:43.729 --> 00:50:46.149 allow us to do it. And on the other hand, we didn't have any sort of hypothesis. 00:50:48.049 --> 00:50:52.449 So yeah, I mean, I guess it speaks again about the representational problem, 00:50:52.709 --> 00:50:56.189 isn't it? But in your mind, was it the end of Tathagamma code or... 00:50:57.353 --> 00:51:03.413 No, no, because we never, we never, we couldn't ever like talk about sequential patterns. 00:51:03.713 --> 00:51:08.293 Okay. And I think that's a critical aspect to talk about, right? 00:51:09.053 --> 00:51:15.053 Yeah, but I, you know, a device such as the brain, like a neural network, 00:51:15.173 --> 00:51:18.213 doesn't necessarily need to treat space and time differently. 00:51:18.333 --> 00:51:22.153 You know, you can convert one into the other and vice versa. 00:51:22.153 --> 00:51:28.913 So, you know, that's, and people do, if you're compressing a movie, 00:51:29.193 --> 00:51:33.993 time's just another dimension along which you're compressing the data alongside 00:51:33.993 --> 00:51:38.633 space, you know, so you're looking for patterns that are temporal and spatial 00:51:38.633 --> 00:51:41.313 within the image and temporal across frames. 00:51:41.313 --> 00:51:45.813 And so you'd think the hippocampus is going to do something similarly clever. 00:51:46.013 --> 00:51:52.093 It's just going to look for ways of compressing information across all these dimensions. 00:51:52.733 --> 00:51:55.893 I don't think time is going to be particularly privileged there. 00:51:56.113 --> 00:52:02.313 And so you could imagine a spatial code for time, which is what the time cells in CA3 may indicate. 00:52:02.613 --> 00:52:08.313 Or you could imagine these phase precession patterns. patterns, 00:52:08.473 --> 00:52:12.313 but I think this is something a little bit naïve to assume that because there's 00:52:12.313 --> 00:52:15.653 a temporal sequence coming in, it's got to have a temporal storage. 00:52:16.533 --> 00:52:19.853 Well, there's order information you might want to conserve. Yeah, 00:52:19.913 --> 00:52:21.333 but you can do that spatially. 00:52:21.473 --> 00:52:24.753 You can map it into a spatial pattern, which has order. 00:52:25.893 --> 00:52:28.673 But then how do you read it out then? 00:52:29.693 --> 00:52:32.593 It has to be in the way it's encoded. There has to be some 00:52:32.593 --> 00:52:35.833 knowledge about how to read it out you know you so you 00:52:35.833 --> 00:52:39.553 have a a cell which has a timestamp but 00:52:39.553 --> 00:52:42.513 it's a you know you or you have you know depending 00:52:42.513 --> 00:52:48.293 on where you are in the in that bit of of of ca3 you're encoding something about 00:52:48.293 --> 00:52:52.053 time in the sequence you know these these these things are totally possible 00:52:52.053 --> 00:52:58.473 you know the the the space encodes time look at the physiology uh off of sweeps 00:52:58.473 --> 00:53:02.133 and and short wave with ripples and phase precession. 00:53:02.473 --> 00:53:06.913 But there you go, that's a way of decoding space, is that you have a ripple, 00:53:07.633 --> 00:53:10.013 and so the ripple gives you your temporal sequence. 00:53:10.233 --> 00:53:14.673 So the encoding, I mean, yeah, I mean, I think we're not disagreeing fundamentally. 00:53:14.953 --> 00:53:21.433 You have both space and time as mechanisms for storing and retrieving information. 00:53:21.833 --> 00:53:27.853 But there isn't any necessary given that temporal information has to be stored 00:53:27.853 --> 00:53:29.213 in a way that's more time. 00:53:29.513 --> 00:53:33.873 Uncomplete, we completely win you on that. I only threw time away, 00:53:33.873 --> 00:53:35.573 halfway in this discussion. 00:53:35.873 --> 00:53:42.393 So you're not necessarily combining time with your item and context information. 00:53:42.833 --> 00:53:45.433 There's no need to do that. Your time worked the whole thing. 00:53:45.473 --> 00:53:46.353 That was my point also earlier. 00:53:46.593 --> 00:53:51.693 You don't want to mentally time travel in real time. It doesn't buy you anything. Yeah, exactly. 00:53:51.993 --> 00:53:54.613 So that's why you also want to segregate from your time. 00:53:55.688 --> 00:53:58.448 Yeah, yeah. To manipulate it, if you will. And that means you don't have to 00:53:58.448 --> 00:54:01.408 privilege time with having a special storage mechanism. 00:54:01.548 --> 00:54:07.568 But that's something else as saying the temporal dynamics of this circuit is 00:54:07.568 --> 00:54:09.788 encoding relationships within items. 00:54:10.468 --> 00:54:13.768 Yeah, yeah. You have to unpack it, and that has to involve time. 00:54:13.768 --> 00:54:15.608 One reason why you want to kick out time. 00:54:15.748 --> 00:54:18.968 Yeah, yeah. So you can exploit the temporal code. Yeah, yeah. That's ambiguous. 00:54:20.048 --> 00:54:25.668 Yeah, yeah, yeah. So, look, Luis, now we dragged you along with this speculation, 00:54:26.668 --> 00:54:33.488 and we see that you still want to believe in the Tathagatma code which is good, 00:54:34.948 --> 00:54:39.888 this is a great idea but so you're as a psychologist you're in this, 00:54:41.548 --> 00:54:46.508 neurophysiology and also if you want the neuropathology of memory which is a 00:54:46.508 --> 00:54:50.988 word that I've left to patients right so if people would like to follow in your 00:54:50.988 --> 00:54:57.108 footsteps in this this complex domain what would be louis law work a lot. 00:54:59.548 --> 00:55:06.888 Work a lot work a lot yes no yeah work around yes and uh be surrounded by uh 00:55:06.888 --> 00:55:12.208 great people i think this is like the element actually um so i did my phd ambassador 00:55:12.208 --> 00:55:13.368 but then i moved to london, 00:55:14.133 --> 00:55:17.193 And, um, of course I learned like, uh, this is like, again, like, 00:55:17.273 --> 00:55:18.613 uh, different ways of learning, right? 00:55:18.693 --> 00:55:21.513 So the two of them gave me completely different stories. But, 00:55:21.533 --> 00:55:26.653 uh, so when I was in London, um, I had this opportunity to be surrounded by 00:55:26.653 --> 00:55:30.573 like people who are amazing and not just because of the, you know, like the sort of papers. 00:55:30.573 --> 00:55:32.953 So in Barcelona they were not so amazing. No, no, no, no, no. 00:55:32.993 --> 00:55:34.173 It was a completely different. 00:55:34.253 --> 00:55:37.593 So in Barcelona what I learned because I had this sort of opportunity to do the PhD with myself. 00:55:38.173 --> 00:55:41.513 I'm sorry. We're kind of, kind of right. And then you learn that walking is 00:55:41.513 --> 00:55:45.513 an important aspect. Well, we know how science goes, right? That's one of these. 00:55:45.853 --> 00:55:48.873 So you have to dedicate a lot of time. But you're right. I mean, 00:55:48.913 --> 00:55:51.893 the academic culture in London, yeah, fantastic. 00:55:52.313 --> 00:55:56.453 And then I just realized... And Sheffield. And Sheffield, I know. UK generally. 00:55:56.973 --> 00:56:00.653 UK generally. Well, it didn't last long, did it? I'm not saying London itself. 00:56:00.693 --> 00:56:01.813 I'm just saying like this sort 00:56:01.813 --> 00:56:05.573 of like a drive that gives you being surrounded by stimulating people. 00:56:06.973 --> 00:56:11.933 And I think actually this is like a critical aspect. When I talk to, 00:56:11.953 --> 00:56:15.553 now I have my students quite, I mean, I came back a few years ago, 00:56:15.593 --> 00:56:19.473 and I really have to express these ideas to them when they don't want to go 00:56:19.473 --> 00:56:20.893 out and do a postdoc for instance. 00:56:20.933 --> 00:56:24.393 I say, look, it's not just, you know, like living your place is that it's an 00:56:24.393 --> 00:56:25.493 essential ingredient of science. 00:56:25.653 --> 00:56:28.933 You need to not just, you know, like survive in another world. 00:56:28.973 --> 00:56:32.733 You have to realize that this is something real or partially real. 00:56:32.973 --> 00:56:39.153 And, and, and I would, so basically, yeah, work a lot and basically. 00:56:39.213 --> 00:56:40.873 Make friends. Put yourself out there. 00:56:41.313 --> 00:56:44.833 But the second thing is, there's something about the relationship between Tony 00:56:44.833 --> 00:56:46.473 and myself that you're not aware of yet. 00:56:47.073 --> 00:56:50.673 Tony likes traveling, right? And I have to sponsor that. 00:56:52.113 --> 00:56:56.753 So always after a podcast, four years later, he visits the lab of the person we interviewed. 00:56:57.053 --> 00:57:03.453 Oh, great. To check when their prediction they made during the podcast is confirmed or not. 00:57:05.293 --> 00:57:09.813 And since in this case, it's just a tram ticket, we can also include your lab 00:57:09.813 --> 00:57:11.853 now. Well, you can come too. 00:57:13.353 --> 00:57:14.853 I might leave the lab. 00:57:16.553 --> 00:57:21.073 So what's the prediction that you really would like to stick your neck out for today? 00:57:22.733 --> 00:57:28.553 For my lab? Or in general? For my work. Yeah, because it turns out to your lab 00:57:28.553 --> 00:57:30.273 to check whether you go fund it or not. 00:57:34.313 --> 00:57:39.573 I kind of have this sort of feeling that the sort of line of research and the 00:57:39.573 --> 00:57:45.213 research that I'm currently doing, which is it's going to incorporate more and 00:57:45.213 --> 00:57:47.993 more, I'd say on the one hand, risky projects, 00:57:48.233 --> 00:57:51.393 because that's something that I actually, I believe that this system in Spain 00:57:51.393 --> 00:57:58.693 allows you to do it, and on the one hand, on the other hand, bring technology, 00:57:58.933 --> 00:58:02.953 more technology, other than, um, you know, like classic imaging studies, 00:58:03.193 --> 00:58:04.933 uh, something slightly more creative. 00:58:05.033 --> 00:58:07.753 I mean, that's why we started using like wearable systems, blah, 00:58:07.873 --> 00:58:09.913 blah, blah, all these sort of things. We are not the only ones, of course. 00:58:14.692 --> 00:58:18.532 And something I'm not sure about, but I think that actually the system is pushing us to do it. 00:58:18.572 --> 00:58:21.812 It's something more applicable, something more into, I wouldn't say industry, 00:58:21.952 --> 00:58:29.372 maybe like for patients, but something like to fully dedicate it to kind of like to solve problems. 00:58:29.912 --> 00:58:33.492 So all these sort of ideas are already in the, the seeds are there because all 00:58:33.492 --> 00:58:35.812 the projects have this sort of elements now these days. 00:58:35.852 --> 00:58:38.892 But I think that these three will be, that would be my wish. 00:58:39.092 --> 00:58:41.712 Yeah, but no, that's not the prediction I was expecting. Okay. 00:58:42.212 --> 00:58:45.392 If I'm going to be alive. I made a different prediction on your prediction. 00:58:45.532 --> 00:58:50.972 I'm going to say, I'm going to prove that all recall depends on gamma range 00:58:50.972 --> 00:58:53.252 responses coupled. Theoretical ones. 00:58:53.812 --> 00:58:57.852 I see. But only above 80 hertz. 00:58:59.332 --> 00:59:03.152 That's it. Science is hypothesis driven. That's what I tell us all the time. 00:59:07.192 --> 00:59:07.632 So, 00:59:12.032 --> 00:59:17.472 in four years, How many years? Except for... Qualified. I want two, that's fine. 00:59:20.152 --> 00:59:26.252 So I think, actually, I need to think a bit more. 00:59:29.172 --> 00:59:31.132 It's the kind of thing you have printed on a T-shirt, right? 00:59:35.472 --> 00:59:36.252 So actually. 00:59:39.252 --> 00:59:45.372 So what I think is actually we will be able to explain many fundamental mechanisms, 00:59:45.512 --> 00:59:48.732 many of them would be applied to real-life situations. 00:59:49.692 --> 00:59:52.192 And that's something Ampetish will do it in two or four years, 00:59:52.852 --> 00:59:53.992 like pattern completion, pattern separation. 00:59:54.352 --> 00:59:58.552 But not just for lab-based experiments, something completely naturalistic. 00:59:59.312 --> 01:00:03.912 Ampetish will be able to do it. Will we be able to read out and replay people's 01:00:03.912 --> 01:00:06.532 episodic memories through fMRI? For instance. 01:00:07.192 --> 01:00:11.272 No, my hippocampus always lies. But now, for instance, what we can do It's one 01:00:11.272 --> 01:00:13.292 of the experiences we wanted to have in our company. 01:00:13.412 --> 01:00:16.932 Now, it's already like the data that we have in collaboration with people that 01:00:16.932 --> 01:00:20.012 have already worked with FMRI, which is fantastic, but at the same time, 01:00:20.092 --> 01:00:24.252 the EEG gives you the possibility to do it in a, let's say, the fine-grained thing. 01:00:24.512 --> 01:00:29.332 We can, if you watch a movie, like standard movie, I don't know if you know the papers. 01:00:29.532 --> 01:00:32.632 Yeah, I know these. It's a bit grainy. 01:00:35.592 --> 01:00:41.272 We did it with the EEG, and it worked fantastically well. I truly believe in this. 01:00:42.192 --> 01:00:48.812 And it's amazing. I mean, you can't simply like disentangle and then look for 01:00:48.812 --> 01:00:50.072 units. What's resolution? 01:00:51.952 --> 01:00:57.272 You're gonna say like, oh, no, no. Okay, the subject is looking at a human being, 01:00:57.352 --> 01:00:59.552 or you're gonna say the subject is looking at, 01:01:01.108 --> 01:01:03.928 a spanish no no no sorry so i was not saying like 01:01:03.928 --> 01:01:06.608 that you're like decoding this sort of information i'm saying 01:01:06.608 --> 01:01:09.368 like when you watch a movie i'm saying that from the 01:01:09.368 --> 01:01:12.148 signal right you're watching a movie yeah watching like our 01:01:12.148 --> 01:01:18.308 interview here and uh right right by didn't ask anything and just by your brain 01:01:18.308 --> 01:01:23.768 signal we can now know which sort of chunks of information are going to be forgotten 01:01:23.768 --> 01:01:29.948 in the later recall okay okay okay right okay so you can tell me afterwards, 01:01:30.168 --> 01:01:32.368 which information the movie didn't care about? 01:01:33.248 --> 01:01:37.128 Maybe. That's maybe not the kind of application I would spend money on. 01:01:38.508 --> 01:01:42.148 But, no, seriously, because I'm... 01:01:44.468 --> 01:01:48.148 Advertising. But that's one small step. 01:01:49.308 --> 01:01:55.588 A small step for Louise, a big step for neuromarketing. That's a real ambition for four years. 01:01:56.268 --> 01:02:00.308 Well, you see, Tony, you better be prepared. Luis Fontanilla, 01:02:00.388 --> 01:02:03.968 thank you very much for this conversation. Thank you very much for the interview. Fantastic. 01:02:05.228 --> 01:02:10.868 The CSN podcast was produced by the Convergent Science Network of Biometrics 01:02:10.868 --> 01:02:17.268 and Biohybrid Systems, a project funded by the European Sevens Research Framework Program. 01:02:19.008 --> 01:02:24.148 For more interviews, recorded lectures, or upcoming conferences in the field 01:02:24.148 --> 01:02:30.388 of biometrics and biohybrid systems, go to csnnetwork.eu. 01:02:30.480 --> 01:02:38.160 Music. 01:02:30.708 --> 01:02:32.588 And thank you for listening.