WEBVTT 00:20.290 --> 00:22.930 This week , we are pleased to host 00:22.930 --> 00:25.600 Doctor Laila Debachi to talk about how 00:25.600 --> 00:27.969 sequential event representations are 00:27.969 --> 00:30.809 formed . We read her recent paper on 00:30.809 --> 00:32.976 the different effects of conscious and 00:32.976 --> 00:35.610 non-conscious memory retrieval on 00:35.610 --> 00:37.650 long-term memory , and this week we 00:37.650 --> 00:39.372 read her classic paper on what 00:39.372 --> 00:41.770 constitutes an episode . An episodic 00:41.770 --> 00:44.103 memory , which focused more on encoding . 00:44.409 --> 00:46.576 Doctor Devaci is currently a professor 00:46.576 --> 00:48.742 of psychology at Columbia University , 00:48.742 --> 00:50.631 where she's leading the charge to 00:50.631 --> 00:53.040 investigate how dynamic experiences are 00:53.040 --> 00:55.689 transformed into lasting memories and 00:56.009 --> 00:58.849 how they update knowledge . She is a 00:58.849 --> 01:01.071 juggernaut , you guys , in the field of 01:01.071 --> 01:03.238 cognitive neuroscience with behavioral 01:03.238 --> 01:05.460 and imaging techniques , making a heavy 01:05.460 --> 01:07.460 impact in the field of learning and 01:07.460 --> 01:09.293 memory . So over to you , Doctor 01:09.293 --> 01:11.516 Devachi , thanks for being here . Thank 01:11.516 --> 01:13.349 you , Kevin , uh , for that kind 01:13.349 --> 01:15.405 introduction . Um , I'll just say at 01:15.405 --> 01:17.627 the outset that I'm , I'm happy to take 01:17.627 --> 01:19.793 any questions , um , during the talk . 01:19.793 --> 01:22.370 Just please feel free to speak up . um , 01:22.449 --> 01:24.560 and Kevin , if you could just monitor 01:24.560 --> 01:26.671 the chat , that'd be great . Just let 01:26.671 --> 01:29.559 me know . Um , All right . So it sounds 01:29.559 --> 01:32.400 like you've read two papers from two 01:32.400 --> 01:34.580 arms of work that are happening in the 01:34.580 --> 01:37.629 lab . Um , and I'm gonna be giving you 01:37.760 --> 01:41.519 um a partly historical but deep dive 01:41.519 --> 01:43.741 into the event representation work that 01:43.741 --> 01:46.279 we've been doing . Um , this is my new 01:46.279 --> 01:48.390 logo for the lab , as you're , you're 01:48.390 --> 01:50.612 looking at now , and this is thought to 01:50.612 --> 01:52.890 like summarize a lot of the work we do . 01:52.890 --> 01:55.680 So this is , um , to me , these little 01:55.680 --> 01:57.919 thin green lines represent like the 01:57.919 --> 01:59.863 myriad things in the world that we 01:59.863 --> 02:01.863 could pay attention to at any given 02:01.863 --> 02:04.086 moment . There's too much information , 02:04.086 --> 02:06.197 um , that there's so much information 02:06.197 --> 02:07.975 that we're always selecting and 02:07.975 --> 02:09.975 prioritizing information . Um , and 02:09.975 --> 02:12.100 that this ring represents the 02:12.100 --> 02:14.369 attentional capture and so there are 02:14.690 --> 02:16.912 certain elements of our experience that 02:16.912 --> 02:19.259 are attended to , um , and I'm gonna 02:19.259 --> 02:21.259 talk about sequentially attended to 02:21.259 --> 02:24.050 today that may form a single memory or 02:24.050 --> 02:26.080 representation . Um , and then that 02:26.080 --> 02:28.789 event , once structured in these dots 02:28.789 --> 02:31.679 goes on to reactivate and replay in 02:31.679 --> 02:33.846 offline states and that's more related 02:33.846 --> 02:36.068 so I have a lot of work that we've done 02:36.068 --> 02:38.160 on postcoding , reactivation , and 02:38.160 --> 02:40.104 replay , including the unconscious 02:40.104 --> 02:42.880 conscious cued work that I think you 02:42.880 --> 02:44.880 guys read that paper on . But today 02:44.880 --> 02:46.880 we're gonna talk about this initial 02:46.880 --> 02:48.936 transformation of memories from this 02:48.936 --> 02:51.800 dynamic mush of experience into sort of 02:51.800 --> 02:53.839 concrete representations . 02:55.710 --> 02:58.369 Um , so another way to think about that 02:58.369 --> 03:00.880 is experience is like a flowing river . 03:01.110 --> 03:03.630 Um , there's some quotes , uh , that I 03:03.630 --> 03:05.519 like . Time is a sort of river of 03:05.519 --> 03:07.869 passing events and strong as its 03:07.869 --> 03:10.036 current , no sooner is a thing brought 03:10.036 --> 03:11.990 to sight than it is swept by and 03:11.990 --> 03:14.157 another takes its place , and this too 03:14.157 --> 03:17.570 will be swept away . Um , but 03:17.570 --> 03:19.681 although experience is like a flowing 03:19.681 --> 03:21.970 river , reflecting back in our memories 03:21.970 --> 03:24.570 is more like a still pool , right ? So 03:24.570 --> 03:26.514 we have a little bit of motion and 03:26.514 --> 03:28.737 information in our memories , but a lot 03:28.737 --> 03:30.959 of it is just the sort of more static , 03:30.959 --> 03:33.181 still pools of water . And so we became 03:33.181 --> 03:35.449 interested in this , um , um , sort of 03:35.449 --> 03:38.000 transformation of the river to these 03:38.000 --> 03:40.410 still pools in our minds , and how can 03:40.410 --> 03:42.410 we understand that transformation . 03:43.380 --> 03:45.759 Another way to ask that question , um , 03:45.770 --> 03:47.970 from the episodic memory literature is 03:47.970 --> 03:50.240 what is an episode of episodic memory . 03:50.490 --> 03:52.712 I had been studying episodic memory and 03:52.712 --> 03:54.879 rodents and monkeys before doing human 03:54.879 --> 03:57.000 work , and we just used the word and 03:57.000 --> 03:59.399 defined it and created single trials 03:59.399 --> 04:01.732 that we would present to people . But I , 04:01.732 --> 04:03.788 you know , became interested in what 04:03.788 --> 04:05.732 are these episodes , where are the 04:05.732 --> 04:08.066 beginnings , the middles , and the ends , 04:08.066 --> 04:10.010 and how does the brain know how to 04:10.010 --> 04:12.177 carve up experience . Um , so in other 04:12.177 --> 04:14.399 words , how do episodic memories emerge 04:14.399 --> 04:16.839 from dynamic ongoing experience ? Um , 04:16.850 --> 04:19.072 and again , when we started this work , 04:19.072 --> 04:21.017 um , that I'm gonna tell you about 04:21.017 --> 04:22.961 today , it was around in 2008 , so 04:22.961 --> 04:25.128 quite some time ago now . The dominant 04:25.128 --> 04:27.294 paradigm in memory research use single 04:27.294 --> 04:29.517 trials , in other words , episodes were 04:29.517 --> 04:31.809 created by us , the experimenters , and 04:31.809 --> 04:34.649 we would test memory for um paired 04:34.649 --> 04:37.089 associates or pictures or words . 04:39.369 --> 04:41.480 So another way to operationalize that 04:41.480 --> 04:43.730 question is when are mnemonic links 04:43.730 --> 04:45.609 formed across sequential 04:45.609 --> 04:47.720 representations such that they become 04:47.720 --> 04:50.369 integrated and represented as a unit or 04:50.369 --> 04:52.809 memory , but also separated from 04:52.809 --> 04:55.031 adjacent episodes . So we think there's 04:55.031 --> 04:57.087 this tension between integration and 04:57.087 --> 04:59.489 separation that we wanted in dynamic 04:59.489 --> 05:01.656 experience that we wanted to capture . 05:01.929 --> 05:03.985 Um , so again , this tension between 05:03.985 --> 05:06.040 discretization , creating units , as 05:06.040 --> 05:08.318 well as integrating within those units . 05:08.750 --> 05:10.730 Um , and our work , I think , is 05:10.730 --> 05:12.563 consistent with there being both 05:12.563 --> 05:14.799 proactive and retroactive mechanisms 05:14.799 --> 05:16.910 that can support the temporal binding 05:16.910 --> 05:20.380 of experiences . Um . And I won't get 05:20.380 --> 05:22.547 into it . This is like a big statement 05:22.547 --> 05:24.713 and there's a lot of data to back that 05:24.713 --> 05:26.713 up . So hopefully by the end , I'll 05:26.713 --> 05:28.991 share , I'll share a part of this . OK , 05:28.991 --> 05:32.320 so , um , since 2008 , we developed a 05:32.320 --> 05:36.269 paradigm that has revealed a new , uh , 05:36.359 --> 05:38.526 behavioral phenomena . So I decided to 05:38.526 --> 05:41.040 name it the EIA Dura Devachi paradigm , 05:41.119 --> 05:44.079 the EDD paradigm . Um , and the basic 05:44.079 --> 05:46.809 paradigm . Um , looks like this 05:46.809 --> 05:49.031 abstractly . So there's a viewer , this 05:49.031 --> 05:51.260 is experience , people are shown a 05:51.260 --> 05:53.500 series of items , and those could be 05:53.500 --> 05:57.100 words , pictures , scenes , um , we've 05:57.100 --> 05:59.267 used , and actually we have sounds now 05:59.267 --> 06:01.489 that we've used various different kinds 06:01.489 --> 06:03.600 of information . And the idea is that 06:03.600 --> 06:05.544 people are attending to individual 06:05.544 --> 06:07.820 items A , B , C , D , E , F , in 06:07.820 --> 06:11.369 sequential order . But um they are 06:11.369 --> 06:14.209 grouped together by some sort of shared 06:14.209 --> 06:16.130 task or goal state . So the color 06:16.130 --> 06:18.352 refers to the task the participants are 06:18.352 --> 06:20.574 doing . So you might be , for example , 06:20.574 --> 06:22.797 looking at objects and deciding whether 06:22.797 --> 06:24.908 they're indoor or outdoor . You might 06:24.908 --> 06:27.019 be , you might be performing the same 06:27.019 --> 06:29.186 task , but the category of information 06:29.186 --> 06:31.408 changes . So what we do is we present a 06:31.408 --> 06:33.463 list of items , the letters . But we 06:33.463 --> 06:35.574 also create this more um kind of more 06:35.574 --> 06:38.559 stable contextual task or goal state to 06:38.559 --> 06:40.670 try to mirror what's happening in the 06:40.670 --> 06:42.892 real world . And when we do this , um , 06:42.892 --> 06:45.003 so let me just give you a little more 06:45.003 --> 06:47.170 details . So these moments of change , 06:47.170 --> 06:49.226 these moments of delta change in the 06:49.226 --> 06:51.337 task are called boundaries . So we'll 06:51.337 --> 06:53.615 refer to them as boundaries , so A , G , 06:53.615 --> 06:56.040 M and S . And what we then do is after 06:56.040 --> 06:58.720 people have studied um a list of items , 06:59.119 --> 07:01.230 we test their memory for the temporal 07:01.230 --> 07:03.286 order or sequential memory for items 07:03.286 --> 07:05.452 from the list . So we might ask people 07:05.452 --> 07:07.508 which , which item did you encounter 07:07.508 --> 07:11.119 first , B and E or K and N . And this 07:11.119 --> 07:13.839 paradigm always controls for the amount 07:13.839 --> 07:16.760 of time that has passed between each of 07:16.760 --> 07:19.559 these pairs at test . But what differs 07:19.559 --> 07:21.559 is either those pairs come from the 07:21.559 --> 07:24.470 same . task or goal state or they 07:24.470 --> 07:27.720 crossed a boundary . There's any 07:27.720 --> 07:29.498 questions , let me know . So we 07:29.498 --> 07:31.331 basically have different ways of 07:31.331 --> 07:33.442 testing temporal memory . And what we 07:33.442 --> 07:35.498 have found in all of these studies , 07:35.498 --> 07:37.387 this is just to name a few really 07:37.387 --> 07:39.498 robust behavioral effect where memory 07:39.498 --> 07:41.720 for the temporal order is significantly 07:41.720 --> 07:44.519 better if two items come from the same 07:44.519 --> 07:46.686 event than if they come from across an 07:46.686 --> 07:48.797 event . And remember , this is , um , 07:48.797 --> 07:51.019 all of these items are equidistant from 07:51.019 --> 07:52.797 each other , so there's nothing 07:52.797 --> 07:54.797 objectively in the environment that 07:54.797 --> 07:58.079 should make that distinct . Do we have 07:58.079 --> 08:00.309 a question ? Yeah , it looks like 08:00.309 --> 08:02.365 Robert has his hand up . Is it OK to 08:02.365 --> 08:05.829 ask a quick question ? How much of this 08:05.839 --> 08:09.709 seems very schematic to me . It would 08:09.709 --> 08:11.950 be across the boundaries you're talking 08:11.950 --> 08:14.117 about , would that potentially be of a 08:14.117 --> 08:16.350 different schema , as opposed to within 08:16.350 --> 08:18.572 the same schema that that would sort of 08:18.572 --> 08:22.339 make sense over . Yeah , so the word 08:22.339 --> 08:25.350 schema has been like overused uh right 08:25.350 --> 08:27.630 now and underdefined in the literature . 08:27.750 --> 08:31.630 So , um , in this work , the 08:31.630 --> 08:34.469 items are all trial unique , and they 08:34.469 --> 08:36.580 come either from the same category of 08:36.580 --> 08:39.640 visual information , or we've had Like 08:39.640 --> 08:41.696 we have a list that I'll show you an 08:41.696 --> 08:43.751 experiment in this talk where all of 08:43.751 --> 08:46.000 the items are objects . So they're from 08:46.000 --> 08:48.111 the same category . They don't belong 08:48.111 --> 08:50.650 to a particular schema . and we'll talk 08:50.650 --> 08:52.872 about what that means later . I don't , 08:52.872 --> 08:54.983 uh , I think that it's been sort of , 08:54.983 --> 08:57.369 um , overused , but the answer is in 08:57.369 --> 08:59.591 this paradigm , the EDD paradigm , that 08:59.591 --> 09:02.500 is not what's driving the effects . Um . 09:03.460 --> 09:05.682 And we can talk more about that later . 09:05.682 --> 09:07.849 I think I have kind of a longer answer 09:07.849 --> 09:09.849 there , um , but actually . Yeah , 09:10.010 --> 09:12.140 sorry , go ahead . I feel like I 09:12.140 --> 09:14.251 interrupted you . I apologize . I was 09:14.251 --> 09:16.418 hoping to interject a quick question . 09:16.418 --> 09:18.640 Um , I wonder if you could talk about , 09:18.640 --> 09:20.859 uh , within between boundary moments , 09:21.020 --> 09:23.131 you know , within a contiguous , uh , 09:23.131 --> 09:25.820 sequence , is the perception of time 09:25.820 --> 09:29.770 linear or non-linear ? No , that's 09:29.820 --> 09:32.153 exactly what's here in this right graph . 09:32.153 --> 09:35.099 So memory , so I don't have perception 09:35.099 --> 09:37.321 for time actually is interesting . I'll 09:37.321 --> 09:39.432 answer that question , but memory for 09:39.432 --> 09:42.299 time . Um , differs . So , objectively , 09:42.340 --> 09:45.070 people are better at the , you know , 09:45.260 --> 09:47.482 at , in , in their temporal memory , so 09:47.482 --> 09:49.482 they know which items came first or 09:49.482 --> 09:51.371 last , so their sequential memory 09:51.500 --> 09:53.611 objectively is better , but they also 09:53.611 --> 09:55.889 their adjustments and their subjective . 09:55.929 --> 09:58.151 Uh , memory assessments . So people are 09:58.151 --> 10:00.207 more likely to say that B and E were 10:00.207 --> 10:02.318 closer together . They were presented 10:02.318 --> 10:04.429 closer together in time , and they're 10:04.429 --> 10:06.262 more likely to say that KNN were 10:06.262 --> 10:08.318 presented further apart in time . So 10:08.318 --> 10:10.318 that these boundaries are producing 10:10.318 --> 10:12.540 these biases and how closely things are 10:12.540 --> 10:14.707 integrated . Now , interestingly , you 10:14.707 --> 10:16.873 asked about perception of boundaries . 10:16.873 --> 10:18.985 It turns out that in the moment , you 10:18.985 --> 10:21.479 perceive the boundaries as taking more . 10:21.809 --> 10:24.031 Um , as taking less time , sorry . If a 10:24.031 --> 10:26.500 list has many , many more boundaries in 10:26.500 --> 10:28.659 it , let's say you're constantly 10:28.659 --> 10:30.603 changing and there's a lot of high 10:30.603 --> 10:32.715 frequency information that feels like 10:32.715 --> 10:34.437 there's a lot of quote unquote 10:34.437 --> 10:36.659 boundaries , there's little stability . 10:36.659 --> 10:38.881 In the moment you perceive that as , as 10:38.881 --> 10:42.010 going by faster , even though in memory , 10:42.219 --> 10:44.052 those memory representations are 10:44.052 --> 10:46.900 further away . Um , and so I have a TED 10:46.900 --> 10:49.233 talk that I did that was all about that , 10:49.233 --> 10:51.233 how the perception and memory are a 10:51.233 --> 10:54.500 little distinct , um . So it's almost 10:54.500 --> 10:57.219 like the analogy of like a boring 10:57.849 --> 11:00.340 Sunday or very relaxed Sunday . You 11:00.340 --> 11:02.562 feel like time is going really slowly , 11:02.562 --> 11:04.729 and later on , you're not gonna have a 11:04.729 --> 11:06.896 lot of memories for that information . 11:06.896 --> 11:08.951 But a really , really busy day where 11:08.951 --> 11:11.118 you're constantly running around , you 11:11.118 --> 11:10.609 feel like time's going really quickly , 11:10.739 --> 11:12.850 later on , you're probably gonna have 11:12.850 --> 11:14.628 more separated memories of that 11:14.628 --> 11:18.429 experience . You don't have to watch my 11:18.429 --> 11:20.596 TED Talk . I don't , I don't , I'm not 11:20.596 --> 11:22.818 particularly proud of that , you put it 11:22.818 --> 11:25.040 in the chat . That's very interesting , 11:25.040 --> 11:28.809 thank you . Yeah . OK . So given 11:28.809 --> 11:31.140 this broad , so this is a really robust 11:31.140 --> 11:33.362 behavioral effect and we've now seen it 11:33.362 --> 11:35.529 other people have replicated it , um . 11:36.130 --> 11:38.090 We hypothesize that the behavioral 11:38.090 --> 11:40.201 effect of increased temporal sequence 11:40.201 --> 11:42.312 memory within events would be related 11:42.312 --> 11:44.034 to some sort of working memory 11:44.034 --> 11:46.239 mechanism that maintains event 11:46.239 --> 11:48.489 representations across stable context . 11:48.570 --> 11:50.890 So the idea is that somewhere in the 11:50.890 --> 11:53.057 brain , there's information about this 11:53.057 --> 11:55.335 blue . Event being a single event , and 11:55.335 --> 11:57.391 the brain is able to incorporate and 11:57.391 --> 12:00.174 encode each of these items distinctly 12:00.174 --> 12:02.585 such that they become better separated 12:02.974 --> 12:05.215 and represented in memory . So we think 12:05.215 --> 12:07.048 of this encoding as some sort of 12:07.048 --> 12:09.104 working memory mechanism . And we're 12:09.104 --> 12:11.326 starting , this is sort of like a shift 12:11.326 --> 12:13.493 to , uh , the kinds of work that we're 12:13.493 --> 12:15.826 doing now and looking at working memory . 12:15.826 --> 12:15.414 But I want to give you some more 12:15.414 --> 12:18.090 details . So , Let's start with the 12:18.090 --> 12:20.146 very first paper . So the very first 12:20.146 --> 12:22.479 paper that we did on this was back , uh , 12:22.479 --> 12:24.590 I think it was published in 2011 . We 12:24.590 --> 12:26.757 had people in an FMRI scanner and they 12:26.757 --> 12:28.868 were reading narratives . And this is 12:28.868 --> 12:31.034 more related to schemas , the question 12:31.034 --> 12:33.090 that Max , I think it's Max with two 12:33.090 --> 12:35.770 X's asked . OK . So people are in the 12:35.770 --> 12:37.937 scanner and they would read a sentence 12:37.937 --> 12:39.992 like he turned on some music to help 12:39.992 --> 12:42.214 him focus on his work . Then there'd be 12:42.214 --> 12:44.270 a delay because we need this an FMRI 12:44.270 --> 12:46.548 scanning because of the whole response . 12:46.548 --> 12:50.150 It's quite sluggish . And then A 12:50.150 --> 12:54.070 boundary sentence contained um Of 12:54.070 --> 12:56.359 temporal memory shifts . So a while 12:56.359 --> 12:58.248 later , he discovered some useful 12:58.248 --> 13:00.359 information and made a few notes . So 13:00.359 --> 13:02.137 this is just a vignette about a 13:02.137 --> 13:04.026 protagonist studying . One of the 13:04.026 --> 13:06.081 sentences implies that more time has 13:06.081 --> 13:08.248 passed . So there's just one word that 13:08.248 --> 13:11.150 is signifying the temporal duration of 13:11.150 --> 13:14.429 this event is , is a while later , and 13:14.429 --> 13:16.596 that we consider a boundary sentence . 13:17.559 --> 13:19.960 Compare that to a moment later , he 13:19.960 --> 13:22.127 discovered some useful information and 13:22.127 --> 13:24.127 made a few notes . So these are our 13:24.127 --> 13:26.349 control sentences . I want to highlight 13:26.349 --> 13:25.640 that here , even though this is kind of 13:25.640 --> 13:27.840 dynamic reading and there's a lot of 13:27.840 --> 13:29.896 information people are maintaining , 13:29.896 --> 13:31.673 the only difference between the 13:31.673 --> 13:33.618 boundary sentences and the control 13:33.618 --> 13:35.784 sentences was one word , implying that 13:35.784 --> 13:37.896 more time had passed or only a moment 13:37.896 --> 13:40.349 had passed . So these are cues to what 13:40.349 --> 13:42.571 whether or not you're moving into a new 13:42.571 --> 13:44.682 event or maintaining in this or being 13:44.682 --> 13:46.405 maintained in the same event . 13:48.780 --> 13:51.260 So thinking about what's happening in 13:51.260 --> 13:55.140 this situation , we um This is just 13:55.140 --> 13:57.196 the experiment flowing over time and 13:57.196 --> 13:59.362 we've sprinkled in a while later and a 13:59.362 --> 14:01.520 moment later sentences . And if those 14:01.520 --> 14:03.679 have an impact , then we expect 14:03.679 --> 14:05.679 people's neural representations and 14:05.679 --> 14:07.568 cognitive representations to look 14:07.568 --> 14:09.457 something like this . So here's a 14:09.457 --> 14:11.457 boundary , here's a memory . He's a 14:11.457 --> 14:13.457 boundary , and here's a memory . We 14:13.457 --> 14:15.568 don't expect the control sentences to 14:15.568 --> 14:17.179 produce um any sort of event 14:17.179 --> 14:19.346 segmentation in memory , but we wanted 14:19.346 --> 14:21.457 to test that to be sure . Um , and so 14:21.457 --> 14:23.179 the way that we did that is we 14:23.179 --> 14:25.123 presented after people read um the 14:25.123 --> 14:27.340 narratives , we presented sentences 14:27.340 --> 14:29.451 that were cues , and we asked them to 14:29.451 --> 14:31.396 recall the very next sentence that 14:31.396 --> 14:34.020 happened in , in the narrative . And so 14:34.020 --> 14:36.380 we included pre-boundary sentences and 14:36.380 --> 14:38.158 asked them to recall across the 14:38.158 --> 14:41.159 boundary . Or boundary sentences and 14:41.159 --> 14:43.440 then they had to recall within the 14:43.440 --> 14:46.880 event . And so we expected better 14:46.989 --> 14:49.479 recall if these were really unitized , 14:49.690 --> 14:51.912 if queued with a boundary sentence than 14:51.912 --> 14:54.079 if queued with a pre-boundary sentence 14:54.079 --> 14:56.246 and behavior . And that's exactly what 14:56.246 --> 14:58.468 we found . So what's being graphed here 14:58.468 --> 15:00.690 is the proportion of total correct cued 15:00.690 --> 15:02.746 recall hits . So the times that they 15:02.746 --> 15:04.912 were able to report what happened next 15:04.912 --> 15:07.079 in the story . Um , I have this little 15:07.079 --> 15:08.912 thing in front of me , um , as a 15:08.912 --> 15:11.134 function of whether they were cued with 15:11.134 --> 15:13.190 the boundary or the pre pre-boundary 15:13.190 --> 15:15.357 sentence . I don't know if I'm getting 15:15.357 --> 15:17.412 ahead of myself , but my question is 15:17.412 --> 15:19.190 about , um , do you ever test a 15:19.190 --> 15:21.301 backwards prediction , so like C to B 15:21.301 --> 15:23.630 within boundary ? Uh , yes , so what we 15:23.630 --> 15:25.909 found in , in this experiment , we 15:25.909 --> 15:28.076 didn't , but in subsequent experiments 15:28.076 --> 15:31.830 we had people free recall . Um , what , 15:32.429 --> 15:34.429 so I should say that when they do , 15:34.429 --> 15:36.540 sorry , when they do make a mistake , 15:36.540 --> 15:38.762 they're more likely to make a mistake , 15:38.762 --> 15:41.190 a mistake that preserves the boundary 15:41.190 --> 15:43.134 information . So they will be more 15:43.134 --> 15:45.134 likely to go backwards and forwards 15:45.134 --> 15:47.190 when they make mistakes . It doesn't 15:47.190 --> 15:49.357 happen very much in these narratives . 15:49.357 --> 15:51.579 Um , vignettes because there's a lot of 15:51.579 --> 15:53.690 inferential thinking that you can use 15:53.690 --> 15:55.559 to not go backwards . But in our 15:55.559 --> 15:58.679 simpler EDD paradigms , what we notice 15:58.679 --> 16:01.119 in free recall is that people will 16:01.119 --> 16:03.286 recall items within the same event and 16:03.286 --> 16:05.452 then when they kind of don't know what 16:05.452 --> 16:07.563 happens next . They're more likely to 16:07.563 --> 16:09.897 shift away and go to another event . Um , 16:09.897 --> 16:11.897 and actually they're more likely to 16:11.897 --> 16:13.897 recall an event boundary , which is 16:13.897 --> 16:13.719 interesting . It's another line of work 16:13.719 --> 16:15.941 we're looking at . So their free recall 16:15.941 --> 16:17.663 behavior really respects these 16:17.663 --> 16:19.890 boundaries as well . Um , but they , 16:20.090 --> 16:22.034 and when they make mistakes , they 16:22.034 --> 16:24.034 don't , they don't cross boundaries 16:24.034 --> 16:26.257 when they do that , they stay within an 16:26.257 --> 16:29.369 event . OK , so , um , 16:30.780 --> 16:32.891 In this narrative , this is the first 16:32.891 --> 16:34.947 experiment where we , where we first 16:34.947 --> 16:36.613 reported this effect where um 16:36.613 --> 16:38.613 sequential memories that are within 16:38.613 --> 16:40.947 events versus across events , and again , 16:40.947 --> 16:40.869 the amount of time is controlled for 16:41.080 --> 16:43.247 here in these studies and all of these 16:43.247 --> 16:45.130 studies . And we refer to this 16:45.130 --> 16:47.297 difference as mnemonic chunking . This 16:47.297 --> 16:49.463 is the extent to which you can call it 16:49.463 --> 16:51.408 a lot of things . You're unitizing 16:51.408 --> 16:53.630 these sequential items into an episodic 16:53.630 --> 16:55.686 memory , and we're going to use this 16:55.686 --> 16:57.686 behavioral measure as an individual 16:57.686 --> 17:00.090 differences measure later on to learn 17:00.090 --> 17:02.201 more about the neural mechanisms that 17:02.201 --> 17:04.479 are critical for the mnemonic chunking . 17:04.479 --> 17:06.479 So the better you are , you're more 17:06.479 --> 17:08.368 likely to integrate within versus 17:08.368 --> 17:11.489 across an event . And so at this time , 17:11.558 --> 17:13.280 we thought there could be many 17:13.280 --> 17:15.336 mechanisms , but we thought about at 17:15.336 --> 17:17.336 least 2 that we could test with the 17:17.336 --> 17:19.280 neuroimaging data . So what neural 17:19.280 --> 17:21.447 mechanisms support episodic chunking ? 17:21.447 --> 17:23.999 So one , it could be two things . Is it 17:23.999 --> 17:26.110 the discretization that's important , 17:26.438 --> 17:29.869 um , or the integration ? And by , um , 17:29.879 --> 17:31.990 and then we could ask how these , and 17:31.990 --> 17:34.046 the way that we're gonna like . Um , 17:34.046 --> 17:36.760 test this is by looking at correlations 17:36.760 --> 17:39.199 between neural correlates of 17:39.199 --> 17:41.366 discretization and integration and ask 17:41.366 --> 17:43.719 which one is a better predictor of the 17:43.719 --> 17:45.775 behavioral measures of chunking . So 17:45.775 --> 17:47.775 we're really gonna use the behavior 17:47.775 --> 17:50.052 here and not just to look at the brain . 17:50.052 --> 17:52.275 Um , so discretization , you've heard a 17:52.275 --> 17:54.330 lot about , uh , you can't pick up a 17:54.330 --> 17:56.219 paper without reading about event 17:56.219 --> 17:58.275 segmentation these days , you know , 17:58.275 --> 18:00.330 Jeff Sachs early beautiful paper and 18:00.330 --> 18:02.552 events segmentation theory with Barbara 18:02.552 --> 18:04.441 Tversky is all about boundaries . 18:04.441 --> 18:06.552 Boundaries is being really critical . 18:06.552 --> 18:08.719 These are moments of change . It makes 18:08.719 --> 18:10.441 sense that people are studying 18:10.441 --> 18:12.552 boundaries a lot because they're easy 18:12.552 --> 18:12.119 to measure . There's usually 18:12.119 --> 18:14.199 attentional fluctuations , novelty 18:14.199 --> 18:16.199 responses of boundaries . So it's 18:16.199 --> 18:18.479 possible that they're really critical 18:18.479 --> 18:20.701 in creating these sequential memories , 18:20.701 --> 18:22.868 that would be sort of a discretization 18:22.868 --> 18:25.020 argument . So , We could look at 18:25.020 --> 18:27.750 boundary responses in the brain and um 18:27.750 --> 18:31.750 identify where we see boundary um 18:31.750 --> 18:34.083 increases compared to control sentences . 18:34.083 --> 18:36.083 So where are these boundary effects 18:36.083 --> 18:38.306 happening ? What parts of the brain are 18:38.306 --> 18:40.028 responding to boundaries . For 18:40.028 --> 18:42.139 measuring integration , again , we're 18:42.139 --> 18:44.083 thinking about this working memory 18:44.083 --> 18:46.361 mechanism and the idea of event models . 18:46.361 --> 18:48.361 So here we argued that in order for 18:48.361 --> 18:50.306 these items to be better sequenced 18:50.306 --> 18:52.194 these sentences here to be better 18:52.194 --> 18:54.306 sequentially integrated , there's got 18:54.306 --> 18:56.250 to be a mechanism that encodes and 18:56.250 --> 18:58.579 maintains the event until the next 18:58.579 --> 19:01.530 boundary , and then it drops down . So 19:01.530 --> 19:03.363 this is more of a working memory 19:03.363 --> 19:05.419 regressor . So we wanted to identify 19:05.419 --> 19:07.609 voxels , whose whose patterns were 19:07.609 --> 19:09.560 consistent with what we call this 19:09.560 --> 19:12.209 integration regressor . So increasing 19:12.209 --> 19:14.320 across an event and dropping an event 19:14.320 --> 19:16.098 boundaries . It's basically the 19:16.098 --> 19:18.320 opposite effect of the discretization . 19:18.320 --> 19:20.487 So we're gonna look for both increases 19:20.487 --> 19:22.209 at boundaries and decreases at 19:22.209 --> 19:23.987 boundaries and ask which one is 19:23.987 --> 19:26.410 predicting better memory within events . 19:28.569 --> 19:30.736 And so I'm just gonna show you some of 19:30.736 --> 19:32.847 the data . So this is just looking at 19:32.847 --> 19:34.791 the , the general , general linear 19:34.791 --> 19:37.020 model . Here are regions that respond 19:37.020 --> 19:38.964 to boundaries pretty strongly . So 19:38.964 --> 19:40.798 these voxels are showing greater 19:40.798 --> 19:42.742 activity at boundaries compared to 19:42.742 --> 19:45.390 other sentences . And what you see is 19:45.390 --> 19:47.557 regions that are involved , you know , 19:47.557 --> 19:49.279 previously been um involved in 19:49.279 --> 19:51.469 attention like the precuneus and parts 19:51.469 --> 19:54.069 of the frontal lobe . And uh were there 19:54.069 --> 19:56.291 any regions ? No one had looked at this 19:56.291 --> 19:58.625 integration regressor before this paper ? 19:58.625 --> 20:00.791 Were there regions that showed ramping 20:00.791 --> 20:02.791 activity within events that dropped 20:02.791 --> 20:04.847 down on event boundaries and what we 20:04.847 --> 20:06.958 see that emerged out of this , GLM um 20:06.958 --> 20:09.180 parametric regressor was regions in the 20:09.180 --> 20:11.125 medial temporal lobe . This is the 20:11.125 --> 20:13.180 interrhinal perirhinal cortex , uh , 20:13.180 --> 20:15.347 parts of the lateral temporal lobe and 20:15.347 --> 20:17.236 ventral medial PFC . So these are 20:17.236 --> 20:19.458 regions whose univaried activity really 20:19.458 --> 20:21.569 tracks the ebb and flow of these , of 20:21.569 --> 20:25.069 these events . And then we can ask 20:25.069 --> 20:27.989 which one predicts memory . Um , so 20:27.989 --> 20:30.989 here what I have graphed here is your 20:30.989 --> 20:33.211 individual differences and within event 20:33.211 --> 20:35.545 binding , this is the mnemonic chunking . 20:35.545 --> 20:37.711 So how , how much , how much better do 20:37.711 --> 20:39.711 you remember within event sentences 20:39.711 --> 20:41.711 versus across event sentences ? And 20:41.711 --> 20:43.822 there's a really strong . correlation 20:43.822 --> 20:45.989 between that and the event regressor , 20:45.989 --> 20:47.711 what I call the sort of linear 20:47.711 --> 20:49.878 regressor in the , so this is actually 20:49.878 --> 20:51.822 the result here , the , so what we 20:51.822 --> 20:54.045 looked at the whole brain and ask where 20:54.239 --> 20:56.959 does activity um correlate with 20:56.959 --> 20:59.310 behavior . And we see that the extent 20:59.310 --> 21:01.254 to which these regions are showing 21:01.254 --> 21:03.254 increases and decreases , um , as I 21:03.254 --> 21:05.032 just defined in the integration 21:05.032 --> 21:07.199 regressor predicts memory . So again , 21:07.199 --> 21:09.421 in the interorhinal perirhinal cortex , 21:09.421 --> 21:11.588 ventral striatum , and medial PFC . So 21:11.588 --> 21:13.810 it looks like activity in these regions 21:13.810 --> 21:15.977 is really tracking the event structure 21:15.977 --> 21:18.032 of the experience and predicting how 21:18.032 --> 21:19.643 well you remember with event 21:19.643 --> 21:21.810 information . Um , and this is just um 21:21.810 --> 21:23.977 data that I took from this NPFC region 21:24.550 --> 21:26.661 to show you the , the strength of the 21:26.661 --> 21:26.390 correlation . 21:31.410 --> 21:34.609 OK . So that boundaries influence the 21:34.609 --> 21:36.770 organization of long-term memory by 21:36.770 --> 21:39.489 arresting or resetting ongoing temporal 21:39.489 --> 21:42.449 integration . So our conclusion here is 21:42.449 --> 21:44.616 not that boundaries are critical , but 21:44.616 --> 21:46.671 that they stop ongoing integration . 21:46.671 --> 21:49.005 There are moments of change , resetting , 21:49.005 --> 21:50.893 they're probably related to theta 21:50.893 --> 21:52.949 oscillations and resetting the theta 21:52.949 --> 21:54.949 oscillation . Um , and this is very 21:54.949 --> 21:57.171 different than what the field have been 21:57.171 --> 21:58.949 thinking about and highlighting 21:58.949 --> 22:01.060 boundaries themselves as being really 22:01.060 --> 22:03.282 critical . Um , boundaries are indeed a 22:03.282 --> 22:03.089 place where there's a lot of different 22:03.089 --> 22:05.422 cognitive operations that are happening , 22:05.422 --> 22:07.700 and I'm happy to talk about that later , 22:07.700 --> 22:09.922 but this , um , this is more consistent 22:09.922 --> 22:11.589 with working memory promoting 22:11.589 --> 22:14.530 integration . Specifically across 22:14.530 --> 22:16.670 participants , ramping activity in 22:16.670 --> 22:19.420 BMPFC and medial temporal lobe cortex 22:19.949 --> 22:22.116 correlates with behavioral measures of 22:22.116 --> 22:24.369 mnemonic chunking . Um , but we were 22:24.369 --> 22:26.480 interested then in what's the role of 22:26.480 --> 22:28.591 the hippocampus . So now we know this 22:28.591 --> 22:30.702 is a structure that's really critical 22:30.702 --> 22:32.869 for memory . It's also really critical 22:32.869 --> 22:35.036 for sequences , spatial memory , there 22:35.036 --> 22:37.258 are place cells here , it didn't emerge 22:37.258 --> 22:39.258 in our whole brain analysis . So we 22:39.258 --> 22:41.425 really wanted to understand what is it 22:41.425 --> 22:43.480 doing , how is it contributing , and 22:43.480 --> 22:45.258 how is it distinct from that in 22:45.258 --> 22:47.859 cortical regions . Does it show ramping 22:47.859 --> 22:50.081 activity or other forms of sequence and 22:50.081 --> 22:53.180 coding within events ? OK , so we're 22:53.180 --> 22:55.890 gonna get to that question . So this 22:55.890 --> 22:57.612 first study involved narrative 22:57.612 --> 23:00.459 comprehension , and we were cognizant 23:00.459 --> 23:02.070 that narrative comprehension 23:02.070 --> 23:04.292 necessitates the ongoing maintenance of 23:04.292 --> 23:06.515 information and then we found this like 23:06.515 --> 23:08.681 working memory maintenance effect . So 23:08.681 --> 23:10.903 we were thinking , well , maybe this is 23:10.903 --> 23:13.070 only specific to reading and narrative 23:13.070 --> 23:15.015 comprehension or things like movie 23:15.015 --> 23:14.619 viewing , maybe working memory is 23:14.619 --> 23:16.508 really critical for understanding 23:16.508 --> 23:18.730 movies , and that's why our results are 23:18.730 --> 23:20.969 pointing towards um a working memory 23:20.969 --> 23:24.560 mechanism . Um , by contrast , everyday 23:24.560 --> 23:26.989 memory for events contains sequential 23:26.989 --> 23:28.933 information that's not necessarily 23:28.933 --> 23:30.680 actively maintained during the 23:30.680 --> 23:32.624 experience . It's not as if you go 23:32.624 --> 23:34.624 through your day constantly telling 23:34.624 --> 23:36.847 yourself a story about what's happening 23:36.847 --> 23:36.359 in the world , right ? You're in the 23:36.359 --> 23:38.880 moment experiencing the world and so we 23:38.880 --> 23:40.991 wanted to ask whether the maintenance 23:40.991 --> 23:43.680 of information um is also more likely 23:43.680 --> 23:46.140 to happen within stable context versus 23:46.140 --> 23:48.469 across event boundaries . And so this 23:48.469 --> 23:50.580 is how I kind of operate . I'm a very 23:50.580 --> 23:52.469 spatial person , so I always draw 23:52.469 --> 23:54.691 pictures about how we're thinking about 23:54.691 --> 23:56.358 things . So , um , ignore the 23:56.358 --> 23:58.580 hippocampus for now , but the idea here 23:58.580 --> 23:58.229 is that there are all these different 23:58.229 --> 24:00.540 granularities of information that we're 24:00.540 --> 24:02.762 encountering . There's something called 24:02.762 --> 24:04.984 temporal context , which is very slowly 24:04.984 --> 24:07.979 changing over time . Layered on top of 24:07.979 --> 24:10.035 that , we have spatial contacts , so 24:10.035 --> 24:11.923 you might , you know , be in your 24:11.923 --> 24:14.146 office for a while and then walk into a 24:14.146 --> 24:16.146 meeting like you may have just done 24:16.146 --> 24:18.201 unless it's on Zoom , so you spatial 24:18.201 --> 24:19.923 contexts which are changing at 24:19.923 --> 24:21.479 different frequencies . And 24:21.479 --> 24:23.590 simultaneously , you have really high 24:23.590 --> 24:25.479 frequency information like people 24:25.479 --> 24:27.701 you're encountering objects , actions , 24:27.701 --> 24:29.923 you're viewing , right ? So there's all 24:29.923 --> 24:32.339 these different Simultaneous , 24:32.800 --> 24:35.790 simultaneous , um , streams of 24:35.790 --> 24:37.846 information that are happening , but 24:37.846 --> 24:40.123 they happen at different granularities , 24:40.123 --> 24:42.290 right ? And so we wanted to try to get 24:42.290 --> 24:44.512 out of the narrative . Work for a while 24:44.512 --> 24:46.689 to understand these mechanisms more 24:46.689 --> 24:49.130 confidently . So we designed studies 24:49.130 --> 24:51.352 that had these , these levels , some of 24:51.352 --> 24:53.574 these levels of information that we can 24:53.574 --> 24:55.574 measure some of the same behavioral 24:55.574 --> 24:57.741 measures and neural measures . And one 24:57.741 --> 24:59.852 thing we thought we would see , which 24:59.852 --> 25:02.599 turned out to be wrong actually , was 25:02.599 --> 25:05.439 that the hippocampus might show more 25:05.439 --> 25:07.969 stable neural patterns within events . 25:08.334 --> 25:10.935 Versus across events and that could be 25:10.935 --> 25:13.214 the basis for forming these sequential 25:13.214 --> 25:15.625 event representations . Um , and I'll 25:15.625 --> 25:18.025 cut to the chase and tell you that 25:18.025 --> 25:20.081 what's happening , what we're seeing 25:20.081 --> 25:22.081 now in our data , in fact , is that 25:22.081 --> 25:23.747 hippocampal patterns are more 25:23.747 --> 25:26.145 dissimilar within events than they are 25:26.145 --> 25:28.367 across events . So there's a little bit 25:28.367 --> 25:30.589 of a puzzle that's emerging and this is 25:30.589 --> 25:32.701 an active area of research , but I'll 25:32.701 --> 25:32.464 also tell you why I think that makes 25:32.464 --> 25:36.380 sense . OK , so in our next 25:36.380 --> 25:39.739 study , we presented um in order to 25:39.739 --> 25:41.961 mimic what we think are these more real 25:41.961 --> 25:44.540 world dynamic changes . Now we 25:44.540 --> 25:47.099 presented individual trials . The 25:47.099 --> 25:49.266 objects were always trial unique , and 25:49.266 --> 25:51.377 people had to attend to the object in 25:51.377 --> 25:53.599 the scene and make a judgment about the 25:53.599 --> 25:55.710 object seen pairing . So they're very 25:55.710 --> 25:57.710 much asked to make a judgment about 25:57.710 --> 25:59.877 each parent associate . But as you can 25:59.877 --> 26:02.869 see here , we created events where 26:02.869 --> 26:05.036 there was some stability in the visual 26:05.036 --> 26:07.829 environment . So these red arrows refer 26:07.829 --> 26:09.940 to trials that we consider boundaries 26:09.940 --> 26:11.829 because these are where the scene 26:11.829 --> 26:13.773 changes . So there are quartets of 26:13.773 --> 26:16.050 items where the scene is stable , um , 26:16.060 --> 26:18.227 and then there are a quartets of items 26:18.227 --> 26:20.282 where there's a boundary item in the 26:20.282 --> 26:22.338 middle . OK , so you can see that we 26:22.338 --> 26:24.479 kind of created this discontinuity in 26:24.479 --> 26:26.646 one stream of visual information while 26:26.646 --> 26:28.868 the other stream is always experiencing 26:28.868 --> 26:30.868 novelty , just to try to have these 26:30.868 --> 26:32.979 different layers of information . And 26:32.979 --> 26:35.239 so we considered these , um , this 26:35.239 --> 26:38.239 quartet as an event , the same event or 26:38.239 --> 26:40.461 within event representations , and this 26:40.461 --> 26:42.628 other quartet includes a boundary , so 26:42.628 --> 26:44.906 it's a cross event representation . Um , 26:44.906 --> 26:46.906 and then after people encountered a 26:46.906 --> 26:49.810 list of items like this . We then 26:49.819 --> 26:52.800 tested their memory for two of all of 26:52.800 --> 26:54.578 the objects and faces that were 26:54.578 --> 26:56.744 encountered . And again , we made them 26:56.744 --> 26:58.856 equidistant from each other . This is 26:58.856 --> 27:01.078 actually a mistake . This should not be 27:01.078 --> 27:03.189 Amy Poehler . I just love her so much 27:03.189 --> 27:02.699 that somehow I made that mistake . It 27:02.699 --> 27:04.810 should have been the 1st item and the 27:04.810 --> 27:06.920 4th item in the event . Um , 27:08.900 --> 27:10.789 Um , see , OK , so I'm gonna stop 27:10.789 --> 27:12.829 reading the chat . So , um , so we 27:12.829 --> 27:14.885 asked people , uh , were these close 27:14.885 --> 27:16.996 together or far apart ? Actually , we 27:16.996 --> 27:19.218 gave them four choices . They could say 27:19.218 --> 27:21.218 whether they were um close , really 27:21.218 --> 27:23.385 close , far or very far . And what you 27:23.385 --> 27:25.329 can see here are the proportion of 27:25.329 --> 27:27.496 responses that the participants gave . 27:27.496 --> 27:29.551 So again , people are more likely to 27:29.551 --> 27:31.496 rate um two objects from that were 27:31.496 --> 27:33.607 paired with the same sequential scene 27:33.607 --> 27:37.359 representation as being uh Really 27:37.369 --> 27:39.989 close or being really close together . 27:40.420 --> 27:43.209 Let me try to move on . I'm so sorry . 27:43.420 --> 27:45.587 See , that was a boundary . Now you're 27:45.587 --> 27:49.199 not gonna remember . I'll uh just read 27:49.199 --> 27:51.532 a couple of these , you're not , I mean , 27:51.532 --> 27:53.477 what you're missing in the chat is 27:53.477 --> 27:55.643 people are commenting on person action 27:55.643 --> 27:57.699 object and this looks similar to the 27:57.699 --> 27:57.680 memory athlete where Chris Baldassano 27:57.680 --> 27:59.902 was in here recently and has seen a lot 27:59.902 --> 28:02.560 of cool overlaps in the the stimuli or 28:02.560 --> 28:05.560 at least the um framework maybe . There 28:05.560 --> 28:07.560 is one question that I , I think is 28:07.560 --> 28:09.616 probably worth voicing here . So Ted 28:09.616 --> 28:11.560 asks , can the same experiments of 28:11.560 --> 28:14.670 ramping happen with spaces , spatial 28:14.670 --> 28:16.614 experience , as well as temporal ? 28:16.680 --> 28:18.013 That's Ted's question . 28:21.369 --> 28:25.219 Yeah , so I think the If 28:25.219 --> 28:27.579 I understand the question , is , would 28:27.579 --> 28:29.801 you see ramping within the same spatial 28:29.801 --> 28:32.500 context ? And my guess is whatever is 28:32.500 --> 28:35.770 the most stable contextual feature . 28:36.689 --> 28:40.020 will then serve as sort of more 28:40.020 --> 28:42.187 backgrounded information and then it's 28:42.187 --> 28:44.409 your attention to . So for example , if 28:44.409 --> 28:46.520 you're in a , in a , on a Zoom call , 28:46.520 --> 28:48.631 you're in the same space . It's gonna 28:48.631 --> 28:50.631 be more the information that you're 28:50.631 --> 28:52.687 encountering on , on the Zoom call , 28:52.687 --> 28:54.853 that's gonna structure your memory for 28:54.853 --> 28:57.020 that event . But then on top of that , 28:57.020 --> 28:59.298 you might also have a spatial boundary . 28:59.298 --> 29:01.520 So one of the big questions is like the 29:01.520 --> 29:01.520 hierarchical nesting of multiple 29:01.520 --> 29:03.742 boundaries . And I think that , I'm not 29:03.742 --> 29:05.853 sure if that's what you're asking . I 29:05.853 --> 29:08.076 think the the short answer is yes . I , 29:08.076 --> 29:10.076 I think that's probably true . Um , 29:11.109 --> 29:13.220 Maybe you wanna , do you wanna ask it 29:13.220 --> 29:15.387 again ? I'm sorry . It just seems very 29:15.387 --> 29:17.331 exciting to me to to realize or to 29:17.331 --> 29:19.387 think about the fact that , um , you 29:19.387 --> 29:21.553 know , these , these temporal things , 29:21.553 --> 29:23.665 it's it's , it's actually not the way 29:23.665 --> 29:23.430 we think . We don't , you know , we 29:23.430 --> 29:25.763 think in stories , we think in episodes , 29:25.763 --> 29:27.986 we think in all these things , and then 29:27.986 --> 29:29.986 we , we , but , but , but in fact , 29:29.986 --> 29:32.152 life is this stream and there and time 29:32.152 --> 29:34.430 is , is part is one of the things that 29:34.430 --> 29:36.390 structures everything . And I just 29:36.390 --> 29:38.670 realized , well , space is also . And 29:38.670 --> 29:40.337 so I bet you can run the same 29:40.337 --> 29:42.559 experiments with , you know , put , put 29:42.559 --> 29:44.726 people in a room and have them look at 29:44.726 --> 29:44.630 a bunch of cars and then say , oh gosh , 29:44.670 --> 29:46.726 we have to move to another , another 29:46.726 --> 29:48.948 office for a few minutes , you're gonna 29:48.948 --> 29:51.226 find that that , that , that becomes a , 29:51.226 --> 29:51.099 you know , an episodic boundary . 29:51.550 --> 29:53.606 Absolutely . Yeah , in fact , one of 29:53.606 --> 29:57.150 one of the most um kind of the One of 29:57.150 --> 30:00.030 the effects that has been shown is that 30:00.310 --> 30:02.421 walking through doors does , like you 30:02.421 --> 30:04.477 just said , does create a boundary . 30:04.477 --> 30:06.643 People have done things even as simple 30:06.643 --> 30:08.754 as like actions that you are engaging 30:08.754 --> 30:11.589 in . So like they had subjects either 30:11.589 --> 30:14.069 study something on the screen , and in 30:14.069 --> 30:16.125 the middle of studying , they got up 30:16.125 --> 30:18.180 and they wiped the screen . They had 30:18.180 --> 30:20.402 them do something that was unrelated to 30:20.402 --> 30:22.569 the task they were and that was enough 30:22.569 --> 30:24.125 to flush the working memory 30:24.125 --> 30:26.402 representations and start a new memory . 30:26.402 --> 30:28.402 I mean , I'm sure that it also will 30:28.402 --> 30:30.569 happen when you change people . If you 30:30.569 --> 30:32.791 change , if you , if you change . If if 30:32.791 --> 30:34.739 if it if it seems meaningful that 30:34.739 --> 30:36.906 you've changed who you're talking to , 30:36.906 --> 30:39.183 I bet that's about also , right ? Yeah , 30:39.183 --> 30:41.406 in the end , this is , this is a game I 30:41.406 --> 30:43.461 always play with myself , is that we 30:43.461 --> 30:45.517 should never get too attached to the 30:45.517 --> 30:47.683 content that we're using in any of our 30:47.683 --> 30:47.410 experiments . And there's nothing 30:47.410 --> 30:49.521 special about movies , by the way , I 30:49.521 --> 30:51.688 don't want anyone tell you that . Then 30:51.688 --> 30:53.910 there is narratives , then there is the 30:53.910 --> 30:56.021 designs that we're doing . One is not 30:56.021 --> 30:57.966 more naturalistic than the other , 30:57.966 --> 31:00.132 unless you're studying people's actual 31:00.132 --> 31:02.132 autobiographical experiences . So I 31:02.132 --> 31:04.479 like to say that it's , don't be too 31:04.479 --> 31:06.590 attached to the content . So it's not 31:06.590 --> 31:09.319 time , space , people . It's try to get 31:09.319 --> 31:11.430 inside your own brain and think about 31:11.430 --> 31:13.097 what is your brain doing . So 31:13.097 --> 31:14.930 ultimately , what's happening at 31:14.930 --> 31:17.041 boundaries is that there's a shift in 31:17.041 --> 31:19.319 the neural processing that's happening . 31:19.319 --> 31:21.375 I think across the whole brain , and 31:21.375 --> 31:23.597 that shift is a boundary , and that can 31:23.597 --> 31:25.763 happen in multiple ways . You can just 31:25.763 --> 31:27.709 get distracted . By your stomach 31:27.709 --> 31:29.876 rumbling , and that's enough . If your 31:29.876 --> 31:32.042 attention shifts to your internal body 31:32.042 --> 31:34.265 states , that will create a boundary as 31:34.265 --> 31:36.431 well . So boundaries are really , um , 31:36.569 --> 31:38.569 you know , they're very equitable . 31:38.569 --> 31:40.680 Like anything can create a boundary . 31:40.680 --> 31:42.625 And I think it's probably a neural 31:42.625 --> 31:44.736 shift that's happening in the brain . 31:44.736 --> 31:44.010 And the question is , well , how big of 31:44.010 --> 31:46.177 a shift do you need ? And maybe that's 31:46.177 --> 31:48.343 not even an interesting question . But 31:48.343 --> 31:50.510 I get a lot of questions about what is 31:50.510 --> 31:52.454 a boundary ? My answer , my answer 31:52.454 --> 31:52.290 usually is anything could be a boundary 31:52.290 --> 31:54.401 as long as you're paying attention to 31:54.401 --> 31:57.979 it . Um , so really , yeah . Was there 31:57.979 --> 32:01.599 one more question ? Doug Yes , 32:01.719 --> 32:05.280 hi , um , I love the way you put it , 32:05.599 --> 32:09.359 boundaries are equitable . And uh 32:09.550 --> 32:13.500 it really reflects well on your early 32:13.589 --> 32:16.989 alma mater with Shank and Company back 32:16.989 --> 32:20.739 in the uh 80s and 90s . The ideas of 32:20.739 --> 32:23.550 scripts and moving from this room to 32:23.550 --> 32:25.883 that room . I really like it , you know , 32:26.270 --> 32:28.949 Ted trigered a good question . Is it 32:28.949 --> 32:31.750 just time and space or all kinds of 32:31.750 --> 32:34.589 different boundary mechanisms in the 32:34.589 --> 32:36.959 brain . Really good talk so far . I 32:36.959 --> 32:39.181 love it . Yeah , thank you , and I love 32:39.181 --> 32:41.126 that you're such a scholar too . I 32:41.126 --> 32:42.126 really , um , 32:45.239 --> 32:47.183 I'm always that's a lot . I really 32:47.183 --> 32:49.239 appreciate that . Uh , formerly from 32:49.239 --> 32:51.350 Columbia a long time ago , myself and 32:51.350 --> 32:54.839 David Waltz . Oh . Center of 32:55.310 --> 32:58.729 cognition that we worked on . Excellent . 32:59.089 --> 33:01.640 Anyway , keep going . This is , this is 33:01.640 --> 33:04.670 wonderful . OK , so , um , again , 33:04.969 --> 33:07.849 the boundaries have shifted our memory 33:07.849 --> 33:09.960 for what goes together , what's being 33:09.960 --> 33:11.793 unitized , you're more likely to 33:11.793 --> 33:13.960 remember with an event representations 33:13.960 --> 33:15.960 as very close together , and you're 33:15.960 --> 33:18.182 more likely to label these across event 33:18.182 --> 33:21.000 uh representations as far or very far . 33:21.479 --> 33:24.380 Um , so this is the result that I 33:24.380 --> 33:26.491 already told you about , but I wanted 33:26.491 --> 33:28.547 to concretize it for you and showing 33:28.547 --> 33:30.547 you the actual stimuli . So in this 33:30.547 --> 33:33.099 study , um , we then , um , sorry , I 33:33.099 --> 33:35.155 changed the wording here . We looked 33:35.155 --> 33:37.155 non-boundary or same event and then 33:37.155 --> 33:39.266 boundary , and now we want to look at 33:39.266 --> 33:41.432 neural similarity . So we're gonna use 33:41.432 --> 33:43.099 uh a form of representational 33:43.099 --> 33:45.155 similarity analysis that you'll hear 33:45.155 --> 33:47.377 about from Niko Krigas Korta . Um , and 33:47.377 --> 33:49.488 his designs , he looks at how similar 33:49.488 --> 33:51.599 is an object to many other objects to 33:51.599 --> 33:53.710 learn a lot about the semantic . Uh , 33:53.710 --> 33:55.821 representational space in the brain . 33:55.821 --> 33:57.543 Here , we've taken that , uh , 33:57.543 --> 33:59.766 technique and applied it , uh , to look 33:59.766 --> 34:01.932 at stability over time . So we look at 34:01.932 --> 34:04.155 the extent to which neural patterns are 34:04.155 --> 34:06.210 similar at the beginning of an event 34:06.210 --> 34:08.432 and at the end of an event as a measure 34:08.432 --> 34:10.655 of pattern stability . I should call it 34:10.655 --> 34:12.655 neural stability or similarity . So 34:12.655 --> 34:14.821 we're gonna compare neural patterns at 34:14.821 --> 34:17.169 trial 1 to trial 4 , both when the same 34:17.169 --> 34:19.199 scene or the same event is being 34:19.199 --> 34:21.421 experienced , as well as when there's a 34:21.421 --> 34:23.729 shift and there's a boundary um um 34:23.729 --> 34:26.899 condition . And we can ask , does 34:26.899 --> 34:30.219 neural similarity across time relate to 34:30.219 --> 34:34.090 mnemonic measures of um Of temporal 34:34.090 --> 34:35.757 memory across time , with the 34:35.757 --> 34:37.479 hypothesis being that the more 34:37.479 --> 34:39.646 stability , the more likely you are to 34:39.646 --> 34:41.534 rate those objects as being close 34:41.534 --> 34:44.059 together or very close together . In 34:44.059 --> 34:46.226 other words , they're part of the same 34:46.226 --> 34:48.281 episodic memory , that was our , our 34:48.281 --> 34:50.929 major focus here . OK , so I'm gonna 34:50.929 --> 34:52.873 show you data from the hippocampus 34:52.873 --> 34:54.540 because we're really , really 34:54.540 --> 34:56.540 interested in what was going on and 34:56.540 --> 34:58.262 what we found was that pattern 34:58.262 --> 35:00.373 similarity or stability across time . 35:00.373 --> 35:03.750 Was related to mnemonic measures of 35:03.750 --> 35:07.340 proximity . So the greater um patterns , 35:07.459 --> 35:09.739 the more similar the first item to the 35:09.739 --> 35:12.500 4th item in the event was , the more 35:12.500 --> 35:14.611 likely you were to rate that as close 35:14.611 --> 35:16.889 or very close together compared to far . 35:16.899 --> 35:18.843 So it looks like pattern stability 35:19.020 --> 35:21.739 relates with this paper related to 35:21.739 --> 35:23.906 temporal , um , subjective measures of 35:23.906 --> 35:26.080 temporal memory . But interestingly , 35:26.120 --> 35:27.898 this was driven by the boundary 35:27.898 --> 35:29.953 condition . So if you look within an 35:29.953 --> 35:33.040 event , there is , um , you know , what 35:33.040 --> 35:35.262 I'm gonna hint at here , but we weren't 35:35.262 --> 35:37.318 paying attention to whether it looks 35:37.318 --> 35:39.318 like there's less stability overall 35:39.318 --> 35:41.318 within event versus across events , 35:41.318 --> 35:43.540 which is not what we had anticipated if 35:43.540 --> 35:45.651 this was a hippocampus alone system . 35:45.651 --> 35:47.596 um , and a lot of people and I was 35:47.596 --> 35:49.484 raised to just focus on one brain 35:49.484 --> 35:51.540 structure and try to figure out what 35:51.540 --> 35:53.762 it's doing . Um , but in fact , what we 35:53.762 --> 35:55.929 found is that the hippocampus seems to 35:55.929 --> 35:58.096 be really critical in stability across 35:58.096 --> 36:00.262 events . So maintaining the uh pattern 36:00.262 --> 36:02.484 stability across an event , despite the 36:02.484 --> 36:04.429 boundary and corruption relates to 36:04.429 --> 36:06.830 mnemonic measures of memory . So the 36:06.830 --> 36:08.886 more stability , the more likely you 36:08.886 --> 36:10.997 are to rate two items across an event 36:10.997 --> 36:14.129 as close together versus far away . But 36:14.129 --> 36:16.500 interestingly , one brain region in the 36:16.500 --> 36:18.333 visual cortex , the left lateral 36:18.333 --> 36:20.444 occipital complex , which is a visual 36:20.444 --> 36:22.667 region that responds to objects as well 36:22.667 --> 36:24.889 as faces , but really more to objects , 36:25.100 --> 36:27.610 showed a main main effect of events , 36:27.739 --> 36:30.860 um . Of eventness . So pattern 36:30.860 --> 36:32.860 stability was greater within events 36:32.860 --> 36:35.082 versus across events , and this was the 36:35.082 --> 36:37.138 only region in the brain that showed 36:37.138 --> 36:39.249 that effect . So if we're looking for 36:39.249 --> 36:41.416 an event representation region in this 36:41.416 --> 36:43.527 design where we're presenting objects 36:43.527 --> 36:43.459 with scenes and faces , it looks like 36:43.459 --> 36:47.020 visual cortex might be um maintaining 36:47.020 --> 36:48.909 that event representation or that 36:48.909 --> 36:51.780 context . And furthermore , stability 36:51.780 --> 36:53.780 within the visual cortex related to 36:53.780 --> 36:56.002 memory as well within an event . So the 36:56.002 --> 36:58.058 more stability , the more likely you 36:58.058 --> 36:59.947 were to rate those items as close 36:59.947 --> 37:01.891 together versus far away . So here 37:01.891 --> 37:04.058 already , we're seeing that we have to 37:04.058 --> 37:06.224 start considering hippocampal cortical 37:06.224 --> 37:08.558 interactions . And that the hippocampus , 37:08.558 --> 37:10.840 while the most critical for memory , is 37:10.840 --> 37:12.896 always communicating with the cortex 37:12.896 --> 37:15.270 and probably using cortical stability 37:15.600 --> 37:18.199 as um some sort of neural measure that 37:18.199 --> 37:21.395 it is attending to and using to Either 37:21.395 --> 37:23.754 form sequential event representations 37:23.754 --> 37:25.754 or not . And this is something that 37:25.754 --> 37:27.976 we're actively working on . I know this 37:27.976 --> 37:29.976 is a very complicated set of , uh , 37:29.976 --> 37:32.198 graphs , but I just wanted to highlight 37:32.198 --> 37:32.014 that we really went in strongly 37:32.014 --> 37:34.514 assuming the hippocampus would show the 37:34.514 --> 37:36.915 effect that visual cortex did . And we 37:36.915 --> 37:39.026 learned something new , which is that 37:39.026 --> 37:41.193 the hippocampus may be really critical 37:41.193 --> 37:43.864 for guiding stability across events , 37:44.235 --> 37:46.291 even when the environment is telling 37:46.291 --> 37:48.457 you not to . And that's something that 37:48.457 --> 37:50.457 humans are really good at . So when 37:50.457 --> 37:52.457 your stomach . Uh , rumbles and you 37:52.457 --> 37:54.513 attend to your internal states , you 37:54.513 --> 37:56.679 remember that there's a test next week 37:56.679 --> 37:56.550 and you really need to attend the 37:56.550 --> 37:58.989 lecture . So you might engage in a 37:58.989 --> 38:01.156 little retrieval mechanism that brings 38:01.156 --> 38:03.322 your brain state back to the state you 38:03.322 --> 38:05.433 were in before . That's what we think 38:05.433 --> 38:07.545 might be going on . That's one of the 38:07.545 --> 38:07.510 retroactive mechanisms that I mentioned . 38:10.790 --> 38:12.512 So what's happening in lateral 38:12.512 --> 38:14.679 occipital complex , is visual region . 38:14.679 --> 38:16.679 Now , if you , if you read anything 38:16.679 --> 38:18.790 about visual cortex , it's thought to 38:18.790 --> 38:20.790 uh really be um a really high freak 38:20.790 --> 38:22.580 it's really respond to visual 38:22.580 --> 38:24.790 information in an on-off manner . It 38:24.790 --> 38:27.110 doesn't really have lingering um 38:27.110 --> 38:29.979 effects , but our data was telling us 38:29.979 --> 38:32.030 that it might , you know , maintain 38:32.030 --> 38:34.159 representation . So , In a side 38:34.159 --> 38:36.048 analysis of this data set that we 38:36.048 --> 38:38.290 published just a few years ago , we 38:38.290 --> 38:42.179 looked at pattern similarity , um , we 38:42.179 --> 38:44.123 broke down that pattern similarity 38:44.123 --> 38:46.290 effect that I showed you and we looked 38:46.290 --> 38:48.401 at whether it was more similar within 38:48.401 --> 38:51.560 an event . So the idea here is If Ello 38:51.560 --> 38:53.899 is processing this cup in the beginning , 38:53.989 --> 38:56.156 we know EO doesn't respond to scenes , 38:56.156 --> 38:58.322 by the way , I'm not going to show you 38:58.322 --> 39:00.433 that data , but this is not driven by 39:00.433 --> 38:59.949 the visual information in the scenes . 38:59.989 --> 39:01.711 That was one of the things the 39:01.711 --> 39:03.933 reviewers asked for in the neuron paper 39:03.933 --> 39:06.156 and that was not there . But we know it 39:06.156 --> 39:08.378 responds to objects . We wanted to test 39:08.378 --> 39:10.156 whether LO representation looks 39:10.156 --> 39:12.322 something like this . So it's encoding 39:12.322 --> 39:14.889 the objects , but it's also maintaining 39:14.889 --> 39:17.111 that cup representation through the end 39:17.111 --> 39:19.111 of the event , and that flushing it 39:19.111 --> 39:21.060 while it's also encoding the other 39:21.060 --> 39:24.719 visual pieces of information . And 39:24.719 --> 39:27.060 so here what I'm showing you is pattern 39:27.060 --> 39:30.860 similarity across time , the first , um , 39:30.870 --> 39:33.092 from the cup to Amy Poehler , noticed I 39:33.092 --> 39:35.790 switched her and now she's 4th , um . 39:36.580 --> 39:38.820 As a function of this is only looking 39:38.820 --> 39:40.876 within events now , as a function of 39:40.876 --> 39:43.139 your mnemonic judgments and indeed when 39:43.139 --> 39:45.306 objects are first , which is the , the 39:45.306 --> 39:47.580 stimulus that Ello really prefers , you 39:47.580 --> 39:49.691 see greater pattern similarity across 39:49.691 --> 39:51.747 time when participants remember them 39:51.747 --> 39:53.858 closer together versus further away , 39:53.858 --> 39:55.858 and you see that more so for object 39:55.858 --> 39:57.969 first events than face first events . 39:57.969 --> 40:00.080 So These first events are more likely 40:00.080 --> 40:02.024 processed in other visual cortical 40:02.024 --> 40:04.024 regions like the FFA . So this is a 40:04.024 --> 40:05.802 little bit of a hint . It was a 40:05.802 --> 40:07.747 sub-analysis of that study . So it 40:07.747 --> 40:09.969 wasn't just designed to look for this , 40:09.969 --> 40:11.802 but it was a hint that there was 40:11.802 --> 40:14.044 lingering maintenance of visual 40:14.044 --> 40:16.364 representations within events that 40:16.364 --> 40:18.364 again was probably flushed at event 40:18.364 --> 40:19.364 boundaries . 40:22.379 --> 40:24.268 The visual cortex shows context , 40:24.669 --> 40:26.780 content specific maintenance of event 40:26.780 --> 40:28.891 representations , unlike what we just 40:28.891 --> 40:32.560 said about the hippocampus . Yeah , OK , 40:32.669 --> 40:34.725 I'm not show , I'm not going to show 40:34.725 --> 40:36.780 you any more data and encoding until 40:36.780 --> 40:38.891 the end of the talk , but I wanted to 40:38.891 --> 40:41.002 shift gears and , and I don't want to 40:41.002 --> 40:40.800 take too long here because I want to be 40:40.800 --> 40:43.250 mindful of the time that we have . But 40:43.679 --> 40:45.735 one of the original questions we had 40:45.735 --> 40:48.320 was , is this a single memory ? 40:48.459 --> 40:50.403 Remember , what is an episode , an 40:50.403 --> 40:52.681 episode memory ? So we're showing that , 40:52.681 --> 40:54.848 yes , behaviorally people are grouping 40:54.848 --> 40:56.570 information and subjective and 40:56.570 --> 40:58.681 objective measures of temporal memory 40:58.681 --> 41:00.792 are being shifted by boundaries , but 41:00.792 --> 41:03.015 is that actually now an episodic memory 41:03.015 --> 41:05.070 when you retrieve it ? So , um , how 41:05.070 --> 41:04.870 would we measure that ? We have to look 41:04.870 --> 41:07.070 at retrieval . So this is a series of 41:07.070 --> 41:09.292 studies that Sarah Dubreu had conducted 41:09.292 --> 41:11.292 in my lab . Again , this is the EDD 41:11.292 --> 41:13.292 paradigm just broken up to show you 41:13.292 --> 41:16.189 that here , now we are presenting 41:16.189 --> 41:18.860 people with objects , as well as faces . 41:18.870 --> 41:21.149 And so a train of faces is , is 41:21.149 --> 41:23.780 considered an event . And then a shift 41:23.780 --> 41:25.947 from faces to objects is considered an 41:25.947 --> 41:27.947 event boundary , and we know that's 41:27.947 --> 41:30.113 gonna cause changes in neural activity 41:30.113 --> 41:32.169 in the visual cortex , and again , I 41:32.169 --> 41:34.280 think that's what boundaries are just 41:34.280 --> 41:37.090 shifts in neural activity . So here's 41:37.090 --> 41:39.312 our , within event representations , we 41:39.312 --> 41:41.479 have a no switch condition . Sorry , I 41:41.479 --> 41:44.169 went too quickly . Um , so anyway , I 41:44.169 --> 41:46.330 know I should . Some of these people 41:46.330 --> 41:48.163 have been canceled and I haven't 41:48.163 --> 41:49.941 changed my slides , so don't be 41:49.941 --> 41:52.219 offended by the faces that are on here . 41:52.219 --> 41:54.639 Um , so in a no-w condition , um , 41:54.729 --> 41:56.896 there would be faces , this would be , 41:56.896 --> 41:59.118 think of this as a face event , right ? 41:59.118 --> 42:01.173 And then a switch condition would be 42:01.173 --> 42:03.396 starting with a face and then switching 42:03.396 --> 42:05.562 to objects and then back to faces . So 42:05.562 --> 42:07.729 there , there's boundaries within this 42:07.729 --> 42:09.979 event . And so after studying a list of 42:09.979 --> 42:12.090 faces and objects , people were shown 42:12.090 --> 42:14.257 two faces on the screen , and they had 42:14.257 --> 42:16.479 to simply say which one came first . So 42:16.479 --> 42:18.500 we tested for temporal memory . And 42:18.500 --> 42:20.444 importantly , they were looking at 42:20.444 --> 42:22.556 either objects or faces on both these 42:22.556 --> 42:24.500 trials . So the visual information 42:24.500 --> 42:26.500 they're getting during retrieval is 42:26.500 --> 42:28.722 exactly the same . And what differed is 42:28.722 --> 42:30.889 the encoding status of these , whether 42:30.889 --> 42:32.889 they were encoded with faces in the 42:32.889 --> 42:34.833 middle , or they were encoded with 42:34.833 --> 42:36.778 objects in the middle . So this is 42:36.778 --> 42:38.778 within event and this is in a cross 42:38.778 --> 42:40.889 event condition . And they were asked 42:40.889 --> 42:43.111 which one was more recent , and what we 42:43.111 --> 42:45.222 found , consistent with all our other 42:45.222 --> 42:47.389 findings is that people were better at 42:47.389 --> 42:49.500 recency judgments for items that were 42:49.500 --> 42:51.611 from the same event versus those that 42:51.611 --> 42:53.611 crossed an event boundary . So that 42:53.611 --> 42:55.833 wasn't surprising . But the question we 42:55.833 --> 42:57.889 wanted to ask here was , does do the 42:57.889 --> 42:59.889 intervening representations help to 42:59.889 --> 43:01.944 bridge the gap ? In other words , is 43:01.944 --> 43:03.889 this now an episodic memory that's 43:03.889 --> 43:06.659 being retrieved before people make the 43:06.659 --> 43:10.550 judgment about recency ? And 43:10.550 --> 43:12.661 so , again , this is just showing you 43:12.661 --> 43:14.717 the list that people were seeing and 43:14.717 --> 43:16.939 this , this is the temporal memory test 43:16.939 --> 43:19.050 that they saw it's just being graphed 43:19.050 --> 43:21.050 in a different way here . And so we 43:21.050 --> 43:20.750 wanted to ask , is there evidence that 43:20.750 --> 43:22.861 intervening representations are being 43:22.861 --> 43:24.972 reactivated during memory judgments ? 43:24.972 --> 43:26.861 In other words , when asked about 43:26.861 --> 43:29.070 Rachel Maddow and Kevin Spacey . Have 43:29.070 --> 43:31.389 you formed an episodic memory such that 43:31.389 --> 43:33.629 I can measure the reinstatement of 43:33.629 --> 43:36.110 these intervening faces ? Is it now a 43:36.110 --> 43:38.166 single memory that is reactivated in 43:38.166 --> 43:40.389 mind in order for you to make this 43:40.389 --> 43:43.600 decision ? So we trained a classifier 43:44.030 --> 43:46.429 um on encoding of faces and objects so 43:46.429 --> 43:48.429 we could tell us , tell us how much 43:48.429 --> 43:50.350 faciness or objectness was being 43:50.350 --> 43:52.572 reinstated during these judgments . And 43:52.572 --> 43:54.794 the critical thing is that perceptually 43:54.794 --> 43:56.628 people are seeing faces , so the 43:56.628 --> 43:58.294 classifier shouldn't show any 43:58.294 --> 44:00.699 difference in face output , for example , 44:00.989 --> 44:02.822 um if it's only sensitive to the 44:02.822 --> 44:04.711 perceptual environment . So we're 44:04.711 --> 44:07.045 really asking the classifier to tell us , 44:07.045 --> 44:09.156 is there a difference in the internal 44:09.156 --> 44:11.378 mnemonic representations that are being 44:11.378 --> 44:13.600 activated . And this is what we found . 44:13.600 --> 44:15.656 So in other words , what we found is 44:15.656 --> 44:18.620 category evidence for faces was um 44:18.620 --> 44:20.820 significantly greater when the two 44:20.820 --> 44:23.899 faces were from a face event . Than if 44:23.899 --> 44:27.179 the two faces came from um two separate 44:27.179 --> 44:29.401 events that were broken up by objects . 44:29.401 --> 44:31.457 In other words , the only reason you 44:31.457 --> 44:33.512 should see differences in classifier 44:33.512 --> 44:35.623 output here is if there's differences 44:35.623 --> 44:37.735 in the retrieved content internally . 44:37.735 --> 44:39.846 And it looks like people are bringing 44:39.846 --> 44:42.068 back to mind not only Rachel Maddow and 44:42.068 --> 44:44.068 Kevin Spacey , but they're bringing 44:44.068 --> 44:46.012 back to mind the inter intervening 44:46.012 --> 44:48.012 faces , um , that were part of that 44:48.012 --> 44:50.179 event or I would say episodic memory . 44:50.239 --> 44:52.350 And Sarah didn't stop there . She was 44:52.350 --> 44:54.461 an incredible scientist , um , and we 44:54.461 --> 44:56.628 wanted to know behaviorally , if it is 44:56.628 --> 44:58.899 true that people are reinstating these 44:58.899 --> 45:02.139 intervening faces , are they primed ? 45:02.179 --> 45:04.123 Are people faster at responding to 45:04.123 --> 45:06.346 those faces when asked about them right 45:06.346 --> 45:08.401 away ? That's a way of understanding 45:08.401 --> 45:10.512 whether they're brought back . So are 45:10.512 --> 45:12.457 they automatically recovered ? And 45:12.457 --> 45:14.623 doesn't influence behavior . So we did 45:14.623 --> 45:16.568 this , she did the same paradigm , 45:16.568 --> 45:18.735 asked people , which was more recent , 45:18.735 --> 45:20.512 and then immediately within 500 45:20.512 --> 45:22.735 milliseconds , ask , just ask people to 45:22.735 --> 45:24.957 say old or new to two faces . And those 45:24.957 --> 45:26.846 faces either came from within the 45:26.846 --> 45:28.679 intervening faces or came from a 45:28.679 --> 45:30.689 preceding face that wasn't in that 45:30.689 --> 45:33.959 event . OK , so this is a way of 45:33.959 --> 45:36.181 priming that something that wasn't seen 45:36.181 --> 45:38.292 on the screen , but neurally we think 45:38.292 --> 45:41.199 it's been reinstated . And basically , 45:41.370 --> 45:43.481 that's what we found . So here , what 45:43.481 --> 45:46.330 I'm plotting is RTs or response time to 45:46.330 --> 45:49.409 say old or new to the intervening face 45:49.409 --> 45:51.576 versus the control phase . And you can 45:51.576 --> 45:54.129 see that people are faster at 45:54.129 --> 45:56.770 responding old if the face was in 45:56.770 --> 45:59.010 between the just tested pair than if it 45:59.010 --> 46:01.177 preceded the just tested pair . And we 46:01.177 --> 46:03.066 didn't see that effect for switch 46:03.066 --> 46:05.010 trials , although it looks like it 46:05.010 --> 46:07.066 might be happening a little bit . So 46:07.066 --> 46:09.066 both neural measures and behavioral 46:09.066 --> 46:11.010 measures argue that boundaries are 46:11.010 --> 46:13.232 creating these sequential memories such 46:13.232 --> 46:15.010 that not only are you better at 46:15.010 --> 46:17.066 temporal memory judgments for them , 46:17.066 --> 46:19.177 but you're actually bringing back the 46:19.177 --> 46:21.399 whole memory together , even though you 46:21.399 --> 46:21.320 don't need to , right ? This is , that 46:21.320 --> 46:22.987 is what a memory is . It gets 46:22.987 --> 46:25.209 reactivated and all of its constituents 46:25.209 --> 46:27.376 get reactivated even if you don't need 46:27.376 --> 46:30.800 that information . I think we have some 46:30.800 --> 46:31.530 questions . 46:34.870 --> 46:37.830 No , we're good . OK . 46:39.479 --> 46:41.812 So how much time do we have ? I'm gonna , 46:41.812 --> 46:44.629 I'm thinking I had one question , um , 46:44.679 --> 46:46.457 which is , um , you're , you're 46:46.457 --> 46:48.959 obviously using , um , stimuli that 46:48.959 --> 46:51.126 people that you think everybody's seen 46:51.126 --> 46:53.989 or or knows , and , um , I'm , I'm , 46:54.120 --> 46:56.287 you know , expecting that you're doing 46:56.287 --> 46:58.287 that because if it was , you know , 46:58.287 --> 47:00.790 faces you've never seen or even , um , 47:00.879 --> 47:03.212 you know , images that that didn't , um , 47:03.212 --> 47:05.268 that weren't things you were already 47:05.268 --> 47:07.268 familiar with , uh , you'd spend so 47:07.268 --> 47:09.323 much time in coding that that itself 47:09.323 --> 47:11.212 would be . Uh , a big part of the 47:11.212 --> 47:13.840 building boundaries . Yeah , exactly . 47:14.050 --> 47:16.530 I mean , I think most of what we were 47:16.530 --> 47:18.863 because we're testing sequential memory , 47:18.863 --> 47:20.863 we really care about these temporal 47:20.863 --> 47:22.308 links across of that item 47:22.308 --> 47:24.474 representations . That we don't , like 47:24.474 --> 47:26.697 you said , we , we're not using stimuli 47:26.697 --> 47:28.252 that people don't have item 47:28.252 --> 47:30.308 representations for yet , right ? So 47:30.308 --> 47:32.530 like . Yeah , but Kevin had this really 47:32.530 --> 47:34.752 cool idea , which is he said he thought 47:34.752 --> 47:36.919 it would work , uh , even if you had a 47:36.919 --> 47:39.219 kaleidoscopic , um , images that were 47:39.219 --> 47:41.590 uh generated just for just to be 47:41.590 --> 47:43.646 stimuli , and I , I just wonder what 47:43.646 --> 47:46.889 you think of that idea . I don't , I 47:46.889 --> 47:48.889 haven't seen any good evidence that 47:48.889 --> 47:51.340 people have any memory , explicit 47:51.340 --> 47:53.290 memory for kaleidoscope images . 47:55.810 --> 47:57.770 There's some it's very implicit 47:57.770 --> 47:59.770 statistical learning paradigms like 47:59.770 --> 48:01.826 Anna Shapiro's work and Nick Brown's 48:01.826 --> 48:04.939 work , they use like um Uh , 48:05.010 --> 48:07.177 fractals , something like that . But , 48:07.177 --> 48:09.177 and I think that the brain can show 48:09.177 --> 48:11.500 that it's building up some sort of 48:11.500 --> 48:14.820 predictive signal . Um , but if you ask 48:14.820 --> 48:17.310 people explicitly , they don't have any 48:17.310 --> 48:20.860 idea what statistical representations 48:20.860 --> 48:23.082 have been encountered together . So you 48:23.082 --> 48:25.249 have to do a lot of training , um , so 48:25.249 --> 48:27.249 that they could name that and , you 48:27.249 --> 48:29.360 know , that , that item and name it , 48:29.360 --> 48:31.416 but at least understand what happens 48:31.416 --> 48:33.638 after it . It's really , really hard to 48:33.638 --> 48:35.693 get temporal memory . It's a very Uh 48:37.669 --> 48:39.891 What's the word ? It's just difficult . 48:39.891 --> 48:42.113 You need those . I , I , I haven't seen 48:42.113 --> 48:44.169 any good evidence that people have . 48:44.169 --> 48:46.447 Which actually goes back to your point , 48:46.447 --> 48:46.429 which is if you don't or what we're 48:46.429 --> 48:48.540 talking about , which is , if it , if 48:48.540 --> 48:50.540 it isn't pre-encoded , it's gonna , 48:50.540 --> 48:52.707 it's gonna take energy to go into code 48:52.707 --> 48:54.485 it . Yeah . If you had a lot of 48:54.485 --> 48:56.429 repetitions though , you could get 48:56.429 --> 48:58.540 there . Like , if it's a kaleidoscope 48:58.540 --> 48:58.360 image that looks like you were 48:58.360 --> 49:00.360 unitizing as a movie , and then the 49:00.360 --> 49:02.304 boundaries were where those images 49:02.304 --> 49:04.249 changed drastically , like changed 49:04.249 --> 49:06.800 color . I'm sure you might have some in 49:07.030 --> 49:09.629 recognition of temporal trajectories , 49:09.669 --> 49:11.891 and you might be able to point out when 49:11.891 --> 49:14.058 there's a violation of that sequence . 49:14.058 --> 49:16.002 You wouldn't be able to explicitly 49:16.002 --> 49:17.891 recall , but you might be able to 49:17.891 --> 49:20.290 predict violations of the motion energy 49:20.290 --> 49:22.550 through that temporal . Yeah , it's a 49:22.550 --> 49:24.717 good question . Which is really cool . 49:24.979 --> 49:26.757 All right , don't don't want to 49:26.757 --> 49:28.868 distract us . So I want to answer the 49:28.868 --> 49:31.090 question of , we do have like kind of 5 49:31.090 --> 49:33.090 minutes left , and Professors , you 49:33.090 --> 49:35.312 know , if you'll stay on the line , I'm 49:35.312 --> 49:37.535 sure people , some people hang around , 49:37.535 --> 49:37.340 but other people might have to go . We 49:37.340 --> 49:39.699 do try to cut things off close to 1 49:39.699 --> 49:42.939 p.m. and , uh , since you're probing 49:42.939 --> 49:45.489 for questions , I , what was in my mind 49:45.489 --> 49:47.656 and then also since you brought up the 49:47.656 --> 49:49.767 word explicit . I fall into this kind 49:49.767 --> 49:51.933 of like , you know , two systems , you 49:51.933 --> 49:54.045 know , the squire taxonomy and so I'm 49:54.045 --> 49:57.820 really curious about uh uh not 49:57.820 --> 50:01.520 uh implicit memories , um . I don't 50:01.520 --> 50:03.687 think this is the right time to really 50:03.687 --> 50:05.742 ask the question , but kind of about 50:05.742 --> 50:07.909 the the prediction and this what I was 50:07.909 --> 50:07.840 saying earlier about forward prediction , 50:07.919 --> 50:10.840 backward prediction , how to everything 50:10.840 --> 50:12.951 you've talked about so far is kind of 50:12.951 --> 50:15.007 episodic memories , right ? But what 50:15.007 --> 50:17.284 about non-hippocampal dependent memory , 50:17.284 --> 50:19.284 um , does that fit in here at all ? 50:19.284 --> 50:21.229 Well , I mean , I think what we're 50:21.229 --> 50:24.199 building is there's a question of 50:25.510 --> 50:27.750 Even what is called a hippocampal 50:27.750 --> 50:30.320 memory involves massive communication 50:30.320 --> 50:32.487 between the hippocampus and cortex , I 50:32.487 --> 50:35.340 think is what , so the um 50:37.489 --> 50:40.050 Implicit associations , yeah , I don't 50:40.050 --> 50:42.272 even know how to begin answering them , 50:42.272 --> 50:44.494 and they're definitely distinct , but I 50:44.494 --> 50:46.494 think this work that I'm presenting 50:46.494 --> 50:48.606 today as well as our post encoding uh 50:48.606 --> 50:50.494 replay work is really pointing to 50:50.649 --> 50:52.816 pretty immediate communication between 50:52.816 --> 50:54.969 the hippocampus and cortex and that I 50:54.969 --> 50:56.358 really think that memory 50:56.358 --> 50:58.679 representations ultimately are 50:58.679 --> 51:01.100 represented across . Brain areas , 51:01.209 --> 51:03.265 which is a very different story than 51:03.265 --> 51:05.489 what we have been learning in cognitive 51:05.489 --> 51:07.689 neuroscience . And , and so we're just 51:07.689 --> 51:09.300 at the tip of the iceberg at 51:09.300 --> 51:11.467 understanding functionality by looking 51:11.467 --> 51:13.489 at specificity of separate brain 51:13.489 --> 51:16.929 regions , but the next phase of Um , I 51:16.929 --> 51:19.090 think growth and innovation in 51:19.090 --> 51:21.209 cognitive neuroscience needs to build 51:21.209 --> 51:23.770 different methods and theories for 51:23.770 --> 51:27.209 looking at , um , one across 51:27.209 --> 51:29.209 regional representation . What does 51:29.209 --> 51:31.153 that even mean ? How would we even 51:31.153 --> 51:33.320 think about that ? And the other is in 51:33.320 --> 51:35.542 terms of memory research , probably the 51:35.542 --> 51:37.598 rule now instead of the exception is 51:37.598 --> 51:39.820 that memory representations once formed 51:39.820 --> 51:42.042 are always changing . They're , they're 51:42.042 --> 51:44.153 dynamically changing , whether that's 51:44.153 --> 51:46.153 because of forgetting or because of 51:46.153 --> 51:48.098 integration into existing semantic 51:48.098 --> 51:50.750 networks and we have very little work 51:50.750 --> 51:53.199 and theories about how to begin to look 51:53.199 --> 51:55.959 at that transformation . Um , and so 51:55.959 --> 51:58.015 those are really exciting areas that 51:58.015 --> 52:00.181 our lab's gonna be getting into , um , 52:00.181 --> 52:03.080 in the future . Laila , is there a way 52:03.080 --> 52:05.030 that we can hook , you know , uh , 52:05.040 --> 52:07.989 connect up with you for A question and 52:07.989 --> 52:09.545 answer after this , Kevin ? 52:12.550 --> 52:14.383 Well , I don't , I don't need to 52:14.383 --> 52:16.494 mediate that . I'll drop her email in 52:16.494 --> 52:18.661 the chat . I'm happy . I would love to 52:18.661 --> 52:18.469 be part of those conversations there 52:18.469 --> 52:20.709 too , so , um , yeah , let's send 52:20.709 --> 52:22.909 follow up emails . Yeah , and I have 52:22.909 --> 52:25.020 more time here too . We can leave , I 52:25.020 --> 52:27.131 don't have to get off right at once , 52:27.131 --> 52:29.131 so if anyone wants to stay a little 52:29.131 --> 52:28.939 after , I'm happy to do that as well . 52:28.989 --> 52:30.711 I just want to be , you know , 52:30.711 --> 52:34.149 respectful of your time . Um , I'm not 52:34.149 --> 52:37.580 gonna tell you about Oh , this is kind 52:37.580 --> 52:39.691 of cool . I'll show you one more data 52:39.691 --> 52:41.913 point before you leave , and then we'll 52:41.913 --> 52:43.969 have questions . So I just wanted to 52:43.969 --> 52:46.191 show you what univariate activity looks 52:46.191 --> 52:48.302 like . This is another version of the 52:48.302 --> 52:48.219 EDD paradigm . We stretched it out to 52:48.219 --> 52:50.489 have more items within an event , all 52:50.489 --> 52:53.010 objects here , and the sounds here 52:53.010 --> 52:56.090 trigger event boundaries . And I just 52:56.090 --> 52:58.201 want to show you what the hippocampus 52:58.201 --> 53:00.312 is doing in , in terms of its overall 53:00.312 --> 53:02.850 activity . Now , these orange circles 53:02.850 --> 53:05.183 are boundaries . This is the whole list . 53:05.183 --> 53:07.770 This is 32 items . And what you can see 53:07.770 --> 53:10.209 is that univariate activity dips at the 53:10.209 --> 53:12.320 boundaries . Remember , this is where 53:12.320 --> 53:14.265 working memory representations are 53:14.265 --> 53:16.487 flushed . So it turns off . And then it 53:16.487 --> 53:18.820 increases within the event . This looks , 53:18.820 --> 53:20.765 you can almost see the integration 53:20.765 --> 53:22.820 regressor that I showed you before , 53:22.820 --> 53:24.931 almost looks like what we saw . So at 53:24.931 --> 53:27.098 boundaries , you get a dip in activity 53:27.098 --> 53:29.060 at the hippocampus , and then it 53:29.060 --> 53:31.116 increases with an event . So you can 53:31.116 --> 53:33.116 see the event structure just in the 53:33.116 --> 53:35.338 univariate responses . And compare that 53:35.338 --> 53:37.504 to the court . Oh , this is CA one , a 53:37.504 --> 53:39.727 region of CA one , showing you a really 53:39.727 --> 53:42.014 nice dip at a boundary and then 53:42.014 --> 53:44.014 increases within an event that then 53:44.014 --> 53:45.958 falls down at the boundary . So it 53:45.958 --> 53:47.736 really looks like what I call a 53:47.736 --> 53:49.792 caterpillar . I like to , like , I'm 53:49.792 --> 53:51.958 gonna make hats for the lab that looks 53:51.958 --> 53:51.655 like this , like it really is sensitive 53:51.655 --> 53:55.040 to structure . And then look at cortex . 53:55.389 --> 53:57.611 Cortex does the opposite . Many regions 53:57.611 --> 53:59.722 of cortex . You see this big increase 53:59.722 --> 54:01.556 at boundaries , um , and then it 54:01.556 --> 54:03.667 decreases within the event . So we're 54:03.667 --> 54:05.629 starting to further explore the 54:05.629 --> 54:07.685 trade-offs between these two and how 54:07.685 --> 54:09.907 they're communicating with each other . 54:09.907 --> 54:09.590 I just wanted to show you these plots 54:09.590 --> 54:11.757 because I think it's really fun to see 54:11.757 --> 54:13.923 the raw , sort of the more raw data um 54:13.923 --> 54:16.257 across these very long , the , you know , 54:16.257 --> 54:18.590 each of these lists is about , you know , 54:18.709 --> 54:21.500 what , 3 minutes long , 3.5 minutes 54:21.500 --> 54:23.667 long , what the brain is doing in real 54:23.667 --> 54:26.689 time . Wow . 54:27.729 --> 54:30.007 Yeah , and now I'll stop . I have more . 54:30.007 --> 54:31.951 I don't know why I thought I could 54:31.951 --> 54:34.118 present more data , but we love data . 54:34.118 --> 54:36.340 It's great . I want to get to Katrina's 54:36.340 --> 54:38.320 question though . Thanks . Yeah , 54:38.439 --> 54:40.661 thanks for sharing this for Lila . It's 54:40.661 --> 54:42.828 so , um , so interesting . So I have a 54:42.828 --> 54:44.883 question about the , the integration 54:44.883 --> 54:46.995 and the ramping activity . I know one 54:46.995 --> 54:49.217 of the reasons you showed . That ramped 54:49.217 --> 54:51.272 up as a boun uh um within a boundary 54:51.272 --> 54:54.439 was VMPSC and I'm curious if you have 54:54.439 --> 54:56.606 thought about sort of the , the notion 54:56.606 --> 54:58.717 of like default mode network activity 54:58.717 --> 55:01.050 increasing as a task persists . So like , 55:01.050 --> 55:03.106 you know , right when you're at that 55:03.106 --> 55:05.328 event boundary in your tasks , which is 55:05.328 --> 55:07.439 you need to really engage , but maybe 55:07.439 --> 55:09.717 after 10 trials of doing the same task , 55:09.717 --> 55:11.919 my attention can kind of Wayne , um , 55:11.929 --> 55:13.818 and that that could be reflecting 55:13.818 --> 55:15.929 increases in regions like the MPFC or 55:15.929 --> 55:17.873 that are more kind of default mode 55:17.873 --> 55:21.270 network um related . Yeah , I mean , I 55:21.270 --> 55:23.437 think the , the default mo network has 55:23.437 --> 55:25.603 an interesting history . So originally 55:25.603 --> 55:28.310 it was defined by not engaging in the 55:28.310 --> 55:30.421 external world , which is what you're 55:30.421 --> 55:32.532 sharing , right ? So like when you're 55:32.532 --> 55:34.810 really focused , you get less activity . 55:34.810 --> 55:34.429 And then when you're kind of mind 55:34.429 --> 55:36.651 wandering or , you know , your eyes are 55:36.651 --> 55:38.651 closed and you're resting , you get 55:38.651 --> 55:41.260 increases in activity . But really now , 55:41.310 --> 55:43.477 I think the field is showing that that 55:43.477 --> 55:45.709 mind wandering is containing really 55:45.709 --> 55:49.270 relevant representations . um , and so 55:49.270 --> 55:51.326 I don't think this is a signature of 55:51.326 --> 55:54.530 mind wandering because the integration , 55:54.629 --> 55:57.110 the extent to which the BMPFC is 55:57.110 --> 55:59.229 following that integration pattern 55:59.229 --> 56:01.350 predicts memory for those events and 56:01.350 --> 56:03.461 sequential memories . That's a really 56:03.461 --> 56:05.517 hard memory test , right ? Um , so I 56:05.517 --> 56:07.572 think the correlations with behavior 56:07.572 --> 56:09.628 argue that it's actually critical in 56:09.628 --> 56:11.906 forming those sequential memories . Um , 56:11.906 --> 56:13.961 but it turns out that as soon as you 56:13.961 --> 56:16.128 repeat anything , the BMPFC is a lot , 56:16.128 --> 56:18.072 default network becomes a lot more 56:18.072 --> 56:21.129 active . So our data suggests that when 56:21.129 --> 56:24.229 you're in a stable context , You don't 56:24.229 --> 56:26.173 need to represent your environment 56:26.173 --> 56:28.285 anymore . You don't need attention to 56:28.285 --> 56:30.396 where you are . Like in my office , I 56:30.396 --> 56:32.451 don't need to know that the window's 56:32.451 --> 56:31.949 over here and the doors over here and 56:31.949 --> 56:34.270 I'm avoiding people . And so then the 56:34.270 --> 56:36.270 default , what's called the default 56:36.270 --> 56:38.492 mode network is actually doing a lot of 56:38.492 --> 56:40.603 the complex thinking and retrieval of 56:40.603 --> 56:42.548 information . So I think it's , it 56:42.548 --> 56:44.600 started off as this like bad name , 56:44.669 --> 56:46.836 like , ah , it's just default , like , 56:46.850 --> 56:49.017 It's when you're not new on anything , 56:49.017 --> 56:51.128 but it turns out that it may actually 56:51.128 --> 56:52.794 be the most critical cortical 56:52.794 --> 56:54.906 architecture that represents a lot of 56:54.906 --> 56:56.850 our accumulated information in the 56:56.850 --> 56:59.017 world and our internal thoughts . Um . 56:59.639 --> 57:01.528 I don't know if that answers your 57:01.528 --> 57:03.695 question fully . There's a lot more to 57:03.695 --> 57:03.270 say about that . I appreciate the 57:03.270 --> 57:05.214 question , but I think that that's 57:05.214 --> 57:07.159 something that we're careful about 57:07.159 --> 57:09.159 looking at . Yeah , and I , I agree 57:09.159 --> 57:11.048 with you in that sense . I wasn't 57:11.048 --> 57:12.992 thinking of it as like they're not 57:12.992 --> 57:15.270 paying attention anymore , but in that , 57:15.270 --> 57:15.139 uh , VMPFC being kind of 57:15.139 --> 57:17.260 self-reflective activity that you're 57:17.260 --> 57:19.659 integrating that information with a 57:19.659 --> 57:21.899 self-representation that's also part of 57:21.899 --> 57:25.739 that episodic , yes , 57:25.860 --> 57:28.027 exactly . And in fact we're working on 57:28.027 --> 57:29.971 a review paper right now . That is 57:29.971 --> 57:32.027 mirroring what you just said . We're 57:32.027 --> 57:34.389 trying to understand , could we think 57:34.389 --> 57:37.000 about how the work on social memory and 57:37.000 --> 57:39.419 self-referential processing and emotion 57:39.419 --> 57:41.909 and the stuff that I do like episodic 57:41.909 --> 57:43.742 memory , how they , they all are 57:43.742 --> 57:45.631 engaging medial PFC and trying to 57:45.631 --> 57:47.520 understand exactly what is that , 57:47.520 --> 57:49.520 what's , what's a more foundational 57:49.520 --> 57:51.520 mechanism that might explain all of 57:51.520 --> 57:53.742 that instead of all these siloed fields 57:53.742 --> 57:55.853 looking at it . But yeah , completely 57:55.853 --> 57:55.629 agree with what you said . 58:05.020 --> 58:07.729 Other questions here while uh Doctor 58:07.729 --> 58:11.379 Devaci is being so kind with this . Uh , 58:11.550 --> 58:15.199 over time here . Laila , this is Max . 58:15.330 --> 58:17.608 Um , is that how I pronounce your name ? 58:18.419 --> 58:21.530 Yes , Leila , yeah , when you're , when 58:21.530 --> 58:23.752 we're done here , if you wanna have a , 58:23.752 --> 58:25.919 I have a research idea to discuss with 58:25.919 --> 58:29.350 you potentially . All right . So I 58:29.350 --> 58:31.669 guess I'll send you an email when this 58:31.669 --> 58:35.530 is done , of course . Sure . 58:35.699 --> 58:38.250 Uh , I , I have a question or two 58:38.250 --> 58:42.010 myself , uh . Leila , um , your , 58:42.090 --> 58:45.040 your slide showing the hippocampal , um , 58:45.409 --> 58:47.689 like cycling where , uh , flushing 58:47.689 --> 58:50.969 occurs . Yeah . Uh , you know , at 58:50.969 --> 58:53.219 event boundaries there , um , I guess , 58:53.290 --> 58:55.290 I mean , I just have some naive 58:55.290 --> 58:57.530 questions about , uh , the roles of 58:57.530 --> 58:59.808 different portions of the brain . Like , 58:59.889 --> 59:01.778 Kevin , you mentioned hippocampal 59:01.778 --> 59:05.429 memories a while ago . And I , is it 59:05.429 --> 59:08.169 the case that the hippocampus is 59:08.169 --> 59:10.939 associated with the narrative 59:10.939 --> 59:14.860 generation ? And the second question I 59:14.860 --> 59:17.899 would have is like the the periods 59:17.899 --> 59:21.870 where the The flushing occurs , um , 59:22.350 --> 59:25.719 is the PFC priming the 59:25.719 --> 59:29.469 hippocampus so that The narrative 59:29.469 --> 59:32.419 generation is updated . Does , does , 59:32.659 --> 59:34.715 do these questions make sense ? Am I 59:34.715 --> 59:38.199 missing anything ? Yeah , no , I think 59:38.199 --> 59:42.100 the , um , your intuitions are , are 59:42.100 --> 59:44.267 quite , I think that that's right . So 59:44.267 --> 59:46.810 we do see that . Um , 59:47.870 --> 59:51.000 Dorsolateral prefrontal cortex . I 59:52.840 --> 59:54.800 It's connectivity with , so it is 59:54.800 --> 59:56.967 mediating hippocampal stability across 59:56.967 --> 59:59.600 events . So there is a , so in other 59:59.600 --> 01:00:03.239 words , there is prefrontal hippocampal 01:00:03.239 --> 01:00:06.520 coordination of what , how much of a 01:00:06.520 --> 01:00:09.699 boundary you're experiencing . Um , in 01:00:09.699 --> 01:00:11.866 other words , like , remember I show , 01:00:11.866 --> 01:00:14.088 I showed you the data , the hippocampal 01:00:14.088 --> 01:00:15.921 stability across events predicts 01:00:15.921 --> 01:00:17.755 temporal memory . That effect is 01:00:17.755 --> 01:00:19.866 correlated with prefrontal activity . 01:00:19.866 --> 01:00:21.810 And that I think is an inferential 01:00:21.810 --> 01:00:23.921 process . So in other words , we know 01:00:23.921 --> 01:00:26.088 what our goals are . If our goal is to 01:00:26.088 --> 01:00:25.899 listen to the whole list or learn the 01:00:25.899 --> 01:00:28.260 whole list , then there's a little and 01:00:28.260 --> 01:00:30.149 the boundaries are almost like an 01:00:30.149 --> 01:00:32.371 annoying little interfering moment that 01:00:32.371 --> 01:00:34.649 just makes it harder for us to do that . 01:00:34.649 --> 01:00:37.149 Mhm . That the , the prefrontal cortex 01:00:37.149 --> 01:00:39.379 can sort of reach down and say , now 01:00:39.379 --> 01:00:41.546 you need , you need to bring back that 01:00:41.546 --> 01:00:43.768 information . It was just flushed , but 01:00:43.768 --> 01:00:45.546 let's retrieve it . And there's 01:00:45.546 --> 01:00:47.657 evidence that these mini retrievals , 01:00:47.657 --> 01:00:49.879 there's other work that shows that even 01:00:49.879 --> 01:00:49.350 retrieving something that's passed for 01:00:49.350 --> 01:00:51.517 3 seconds involves the hippocampus and 01:00:51.517 --> 01:00:53.750 prefrontal cortex . So it doesn't have 01:00:53.750 --> 01:00:55.917 to be long . So I think that there are 01:00:55.917 --> 01:00:57.861 these retro , that's a retroactive 01:00:57.861 --> 01:00:59.806 mechanism for saying , get back on 01:00:59.806 --> 01:01:02.139 track . Like , what were you just doing ? 01:01:02.139 --> 01:01:04.306 Um I guess like that generates another 01:01:04.306 --> 01:01:06.417 question in me . When uh the PFC does 01:01:06.417 --> 01:01:08.583 this mediation across event boundaries 01:01:08.583 --> 01:01:10.750 in the hippocampus , um , is there any 01:01:10.750 --> 01:01:12.972 informa any , do we know anything about 01:01:12.972 --> 01:01:15.028 the nature of the information that's 01:01:15.028 --> 01:01:17.083 conveyed across the boundary ? Is it 01:01:17.083 --> 01:01:19.306 more semantic or is it still episodic , 01:01:19.600 --> 01:01:22.120 given that , uh , do you have any sense 01:01:22.120 --> 01:01:26.110 about that ? That's a good question . 01:01:26.340 --> 01:01:28.507 I bet you could design a study to make 01:01:28.507 --> 01:01:30.840 it either . Like , if you , for example , 01:01:30.840 --> 01:01:33.449 if what was relevant is , if everything 01:01:33.449 --> 01:01:35.780 was like semantically related before a 01:01:35.780 --> 01:01:38.113 boundary , like all musical instruments , 01:01:38.113 --> 01:01:40.770 and then you pass over into a beach 01:01:40.770 --> 01:01:42.992 related , you know , let's pack for the 01:01:42.992 --> 01:01:45.214 beach thing . I bet that you , it could 01:01:45.214 --> 01:01:47.270 be more like a let's just activate , 01:01:47.270 --> 01:01:49.270 you know . I need to remember which 01:01:49.270 --> 01:01:51.103 instruments to pack . OK , let's 01:01:51.103 --> 01:01:53.326 activate all the instruments , and then 01:01:53.326 --> 01:01:55.437 you get episodic details . So , so to 01:01:55.437 --> 01:01:57.840 me , the semantic information is always 01:01:57.840 --> 01:01:59.840 intertwined with episodic , and you 01:01:59.840 --> 01:02:02.062 might reactivate one to get the other , 01:02:02.062 --> 01:02:04.110 and it goes vice versa . So I think 01:02:04.110 --> 01:02:06.277 that it could be , it could probably , 01:02:06.277 --> 01:02:08.221 it's probably both , and you could 01:02:08.221 --> 01:02:10.388 design a study to show like one or the 01:02:10.388 --> 01:02:10.040 other . I don't think it's going to be 01:02:10.040 --> 01:02:13.780 limited to one . So the context for my 01:02:13.780 --> 01:02:16.540 questions are that I'm an engineer and 01:02:16.540 --> 01:02:19.500 I'm working to like using these uh 01:02:19.500 --> 01:02:22.260 diffuser diffusion-based generative 01:02:22.260 --> 01:02:25.909 modeling , um . Uh , to , to 01:02:26.280 --> 01:02:29.989 like sample trajectories over time . Um , 01:02:30.479 --> 01:02:32.368 I'm working in that field and I'm 01:02:32.368 --> 01:02:34.535 wondering , you know , how , how can I 01:02:34.535 --> 01:02:36.800 apply this knowledge to , uh , answer 01:02:36.800 --> 01:02:38.856 questions in , in , in the field I'm 01:02:38.856 --> 01:02:41.078 working . So , I mean , that's , that's 01:02:41.078 --> 01:02:43.244 sort of some context . Yeah . I mean , 01:02:43.244 --> 01:02:45.078 I think with humans and with any 01:02:45.078 --> 01:02:47.820 namable stimuli that we use , like 01:02:47.820 --> 01:02:50.042 there's semantic representations or are 01:02:50.042 --> 01:02:52.264 gonna have like a primacy and be really 01:02:52.264 --> 01:02:54.487 important in bringing back any episodic 01:02:54.487 --> 01:02:56.598 details . So if I had to bet money on 01:02:56.598 --> 01:02:59.000 one . I would , I would definitely 01:02:59.000 --> 01:03:00.778 include semantics . It's really 01:03:00.778 --> 01:03:02.889 critical for all the memories that we 01:03:02.889 --> 01:03:04.899 form . Um , semantics are really , 01:03:05.100 --> 01:03:07.211 really important , even though I talk 01:03:07.211 --> 01:03:09.211 about episodic memory is how do the 01:03:09.211 --> 01:03:11.489 specific semantics get pieced together . 01:03:11.800 --> 01:03:14.419 That day versus , you know , today 01:03:14.419 --> 01:03:16.770 tomorrow , right ? It's like sort of 01:03:16.770 --> 01:03:20.280 jumbling them up , um . Anyway , 01:03:21.399 --> 01:03:23.232 I have a question . I think this 01:03:23.232 --> 01:03:25.010 somewhat follows Bob's previous 01:03:25.010 --> 01:03:27.010 question , and it came up initially 01:03:27.010 --> 01:03:29.121 when , um , Ted had asked some during 01:03:29.121 --> 01:03:30.955 your talk , and I , you had said 01:03:30.955 --> 01:03:33.177 something . It's kind of offhand , so I 01:03:33.177 --> 01:03:32.719 want to make sure I caught it was 01:03:32.719 --> 01:03:34.739 something like , um , there's no 01:03:35.340 --> 01:03:37.451 privileged information , anything can 01:03:37.451 --> 01:03:39.618 be the event boundary . I , you know , 01:03:39.618 --> 01:03:41.618 it can , and it was in the , on the 01:03:41.618 --> 01:03:43.840 slide where we had person action object 01:03:43.840 --> 01:03:46.540 and um . You know , one thing that I , 01:03:46.580 --> 01:03:48.770 I was thinking about though is , is , 01:03:48.879 --> 01:03:52.340 um , space , and , and you had said 01:03:52.340 --> 01:03:54.451 something like , well , let's look at 01:03:54.451 --> 01:03:56.673 the brain and let's see what's going on 01:03:56.673 --> 01:03:58.896 in the brain , and , and space seems to 01:03:58.896 --> 01:04:00.896 be represented kind of separately , 01:04:00.896 --> 01:04:02.896 right ? And so does that Is that is 01:04:02.896 --> 01:04:05.007 that privileged information , right ? 01:04:05.007 --> 01:04:07.284 Or is it everything that , so , I mean , 01:04:07.284 --> 01:04:10.350 I guess like , You know , not 01:04:10.350 --> 01:04:12.572 necessarily a particular location , but 01:04:12.572 --> 01:04:14.794 just a spatial representation providing 01:04:14.794 --> 01:04:17.510 this kind of like context while all the 01:04:17.510 --> 01:04:19.399 content's being filled in by this 01:04:19.399 --> 01:04:21.510 person , action , object , particular 01:04:21.510 --> 01:04:23.677 location . So the privilege , there is 01:04:23.677 --> 01:04:25.843 privilege information being like space 01:04:25.879 --> 01:04:29.129 providing the context . Or maybe not . 01:04:29.570 --> 01:04:32.439 Yeah , so space does have 01:04:33.070 --> 01:04:36.479 primacy with long-term memory . So if 01:04:36.479 --> 01:04:38.649 you , if I ask you what did you have 01:04:38.649 --> 01:04:41.590 for breakfast yesterday ? What your 01:04:41.590 --> 01:04:44.110 brain does is immediately tries to 01:04:44.110 --> 01:04:46.277 figure out where you were , even if it 01:04:46.277 --> 01:04:48.499 has nothing to , like , I mean , yeah , 01:04:48.499 --> 01:04:50.721 it has something to do with the kitchen 01:04:50.721 --> 01:04:50.689 or someone else's kitchen or the coffee 01:04:50.689 --> 01:04:52.750 shop or your office , right ? If you 01:04:52.750 --> 01:04:54.972 can retrieve the location , you're more 01:04:54.972 --> 01:04:57.139 likely to retrieve the other details , 01:04:57.139 --> 01:04:59.194 the person , objects , and actions . 01:04:59.194 --> 01:05:01.417 That's been shown behaviorally . But in 01:05:01.417 --> 01:05:03.583 these designs , I think that it really 01:05:03.583 --> 01:05:05.583 is about attention . Um , and so we 01:05:05.583 --> 01:05:07.417 know that space isn't everything 01:05:07.417 --> 01:05:09.583 because you can form episodic memories 01:05:09.583 --> 01:05:11.806 even when you're not leaving the room . 01:05:11.806 --> 01:05:14.020 But what COVID also taught us is that 01:05:14.060 --> 01:05:16.500 that's really good . Like after a long 01:05:16.500 --> 01:05:19.820 time of no spatial just no change in 01:05:19.820 --> 01:05:22.580 context , if you cha if your context is 01:05:22.580 --> 01:05:24.691 always the same , then it becomes not 01:05:24.691 --> 01:05:26.691 diagnostic . And what happened over 01:05:26.691 --> 01:05:28.802 COVID is that people started to think 01:05:28.802 --> 01:05:30.699 they were all losing their memory 01:05:30.699 --> 01:05:33.032 because you couldn't remember what time , 01:05:33.032 --> 01:05:35.366 when something happened , what happened . 01:05:35.366 --> 01:05:37.366 All of our episodic memories became 01:05:37.366 --> 01:05:39.255 completely untethered because our 01:05:39.255 --> 01:05:41.477 contexts never changed . And that was a 01:05:41.477 --> 01:05:43.532 really interesting moment for memory 01:05:43.532 --> 01:05:45.588 research . And so we started , we're 01:05:45.588 --> 01:05:47.643 starting to do some autobiographical 01:05:47.643 --> 01:05:47.399 memory work where we're tracking 01:05:47.959 --> 01:05:50.126 people's experiences and we're looking 01:05:50.126 --> 01:05:52.719 at spatial novelty , but also social 01:05:52.719 --> 01:05:55.679 novelty , emotion , exercise , and what 01:05:55.679 --> 01:05:58.320 we're learning is that the more . 01:05:59.379 --> 01:06:01.620 Event boundaries , the more distinct 01:06:01.620 --> 01:06:03.620 context that you're in in the day , 01:06:03.620 --> 01:06:05.731 predicts better memory for the events 01:06:05.731 --> 01:06:07.842 that day . And it does so for all the 01:06:07.842 --> 01:06:09.953 intervening events . So we're , we're 01:06:09.953 --> 01:06:12.064 starting a new line of work for which 01:06:12.064 --> 01:06:13.787 I'm trying to get funding . So 01:06:13.787 --> 01:06:15.620 hopefully , it's really exciting 01:06:15.620 --> 01:06:17.787 because it's really in the wild . It's 01:06:17.787 --> 01:06:19.953 actually naturalistic . It's not movie 01:06:19.953 --> 01:06:19.860 viewing . People are moving in the 01:06:19.860 --> 01:06:22.027 world and they're recording their real 01:06:22.027 --> 01:06:24.249 autobiographical memory experiences and 01:06:24.249 --> 01:06:25.860 we're really seeing . Pretty 01:06:25.860 --> 01:06:28.385 significant effects of novelty and 01:06:28.385 --> 01:06:30.465 emotion on the day and memory for 01:06:30.465 --> 01:06:32.576 information that day that thinks that 01:06:32.576 --> 01:06:34.864 that is consistent with this idea that 01:06:34.864 --> 01:06:36.808 moving through the world , we need 01:06:36.808 --> 01:06:38.697 multiple contacts and that really 01:06:38.697 --> 01:06:40.808 tethers our , our memory and how well 01:06:40.808 --> 01:06:44.129 we can recall those memories . Um , 01:06:44.209 --> 01:06:47.399 Lila , um , this , this , uh , 01:06:47.889 --> 01:06:50.409 primacy of space , um , I'm wondering 01:06:50.409 --> 01:06:52.576 about relationship , you know , if you 01:06:52.576 --> 01:06:54.576 were , you had a meeting with , you 01:06:54.576 --> 01:06:56.631 know , 5 amazing graduate students , 01:06:56.631 --> 01:06:58.929 one after the other , um , you're , 01:06:59.050 --> 01:07:01.106 you're , you know , how good is that 01:07:01.106 --> 01:07:03.449 relative to going to 3 to 5 different 01:07:03.449 --> 01:07:06.439 offices to talk to people , um , at , 01:07:06.449 --> 01:07:09.370 at , at being boundaries and and being 01:07:09.370 --> 01:07:11.203 uh significant for your memory , 01:07:11.409 --> 01:07:15.100 including . I would still have to 01:07:15.100 --> 01:07:17.979 bet that it would depend , OK , it 01:07:17.979 --> 01:07:19.868 would depend on they'd have to be 01:07:19.868 --> 01:07:21.812 equally arousing and interesting . 01:07:22.330 --> 01:07:24.441 Right ? So people are inherently more 01:07:24.441 --> 01:07:26.608 interesting . So they'd be really give 01:07:26.608 --> 01:07:28.830 space a run for their money . But space 01:07:28.830 --> 01:07:31.163 is weirdly we use space really , really , 01:07:31.163 --> 01:07:33.163 so you would , you would definitely 01:07:33.163 --> 01:07:34.941 remember that you had been in 3 01:07:34.941 --> 01:07:37.219 different offices . But you , you know , 01:07:37.219 --> 01:07:39.441 you might not remember what happened to 01:07:39.441 --> 01:07:39.050 them as much . So I don't know , space 01:07:39.050 --> 01:07:41.328 is so interesting because , by the way , 01:07:41.328 --> 01:07:43.217 we don't really know how space is 01:07:43.217 --> 01:07:45.328 represented at all . All this work is 01:07:45.328 --> 01:07:47.494 going into space . There's place cells 01:07:47.494 --> 01:07:49.439 and hippocampus , there's parietal 01:07:49.439 --> 01:07:51.606 cortex and spatial attention . There's 01:07:51.606 --> 01:07:53.717 parahippocampal place area and seeing 01:07:53.717 --> 01:07:55.828 responsivity . The whole brain and we 01:07:55.828 --> 01:07:57.939 yeah we still don't really understand 01:07:57.939 --> 01:07:59.883 spatial representations , which is 01:07:59.883 --> 01:08:01.828 Telling us that we're not probably 01:08:01.828 --> 01:08:03.828 approaching it in the right way . I 01:08:03.828 --> 01:08:06.050 wonder if it's like that we just have , 01:08:06.050 --> 01:08:05.330 again , we need these better methods 01:08:05.330 --> 01:08:07.929 and theories for approaching our data . 01:08:08.050 --> 01:08:10.106 I just wanted to say that because we 01:08:10.106 --> 01:08:12.106 think we understand space , but the 01:08:12.106 --> 01:08:14.328 whole brain is sensitive to space , and 01:08:14.328 --> 01:08:16.272 we don't really still have a great 01:08:16.272 --> 01:08:18.439 understanding of it . But it's , we do 01:08:18.439 --> 01:08:20.439 use it from memory , whether or not 01:08:20.439 --> 01:08:22.772 that's good or bad , it does seem to be . 01:08:22.772 --> 01:08:25.250 Um , something that we tend to do , and 01:08:25.250 --> 01:08:27.306 actually we have a data point that I 01:08:27.306 --> 01:08:29.528 didn't share with you , but still we're 01:08:29.528 --> 01:08:31.528 writing the paper now . When we ask 01:08:31.528 --> 01:08:33.528 people in these designs to retrieve 01:08:33.528 --> 01:08:36.918 these within event representations . We 01:08:36.918 --> 01:08:38.751 can do something called encoding 01:08:38.751 --> 01:08:40.862 retrieval similarity . We can look at 01:08:40.862 --> 01:08:43.029 how similar is your brain state during 01:08:43.029 --> 01:08:46.338 the retrieval to when you encoded that 01:08:46.338 --> 01:08:48.999 information . And memory memory 01:08:48.999 --> 01:08:51.166 research has shown that the similarity 01:08:51.166 --> 01:08:53.166 between the encoding moment and the 01:08:53.166 --> 01:08:54.999 retrieval moment predicts memory 01:08:54.999 --> 01:08:57.221 success . We've shown that . What we're 01:08:57.221 --> 01:08:59.332 seeing in these event studies is that 01:08:59.332 --> 01:09:01.443 when you're retrieving , I'm pointing 01:09:01.443 --> 01:09:03.499 to this thing , when you're asked to 01:09:03.499 --> 01:09:05.555 retrieve the temporal order of these 01:09:05.555 --> 01:09:07.666 two items . And you get it right , we 01:09:07.666 --> 01:09:09.610 see greater similarity between the 01:09:09.610 --> 01:09:11.777 retrieval trial and its own boundary . 01:09:11.830 --> 01:09:13.830 So it looks like you're reinstating 01:09:13.830 --> 01:09:16.052 that boundary to recall the information 01:09:16.052 --> 01:09:18.108 in the event , which is a circuitous 01:09:18.108 --> 01:09:20.163 route to the memory . You don't have 01:09:20.163 --> 01:09:22.274 direct access . You're not just going 01:09:22.274 --> 01:09:24.163 right to this item and activating 01:09:24.163 --> 01:09:26.330 things around it . You're going to its 01:09:26.330 --> 01:09:28.608 boundary and then unlocking the memory . 01:09:28.608 --> 01:09:30.774 And this , this is really exciting and 01:09:30.774 --> 01:09:32.830 new data , but How does that link to 01:09:32.830 --> 01:09:34.719 space ? It might be , it might be 01:09:34.719 --> 01:09:37.052 similar to the idea of where , you know , 01:09:37.052 --> 01:09:36.259 when's the last time you had a 01:09:36.259 --> 01:09:38.481 hamburger ? Well , you've got to recall 01:09:38.481 --> 01:09:40.815 space or what do you have for breakfast . 01:09:40.815 --> 01:09:43.037 You might go to the boundary moment and 01:09:43.037 --> 01:09:45.148 that boundary moment is often defined 01:09:45.148 --> 01:09:47.148 by a new location . Right ? It just 01:09:47.148 --> 01:09:49.259 often is . It doesn't have to be . It 01:09:49.259 --> 01:09:51.426 could be an emotion too . Emotions are 01:09:51.426 --> 01:09:53.426 really like really big emotions are 01:09:53.426 --> 01:09:55.537 also big boundaries . Um , anyway , I 01:09:55.537 --> 01:09:57.648 wanted to share another piece of data 01:09:57.648 --> 01:09:57.548 with you that I think is pretty cool . 01:09:58.799 --> 01:10:00.855 That is so fascinating . I could sit 01:10:00.855 --> 01:10:02.910 here and talk about memory all day , 01:10:02.910 --> 01:10:04.910 and you guys all should . I'm gonna 01:10:04.910 --> 01:10:07.021 stop the recording . I've got another 01:10:07.021 --> 01:10:06.979 thing I got to get ready for , but guys , 01:10:07.100 --> 01:10:09.322 feel free to keep talking , but be nice 01:10:09.322 --> 01:10:11.520 to uh our guest here . Thank you , 01:10:11.649 --> 01:10:12.720 Professor , for coming in .