stoked to announce i'll be starting my lab at USC jan 2025!
please RT (here or IRL): i'll be looking for a research specialist/lab manager & grad students (fall 2025); email me! (will also be giving a talk at APS tomorrow)
this took entire villages...
Wonder what 2020 did to your memory or sense of time?
🚨 Now out in
@PNASNews
🚨 we tracked how collective events and individuals’ emotional state (sampled throughout 2020) shaped spontaneous autobiographical memory across 3 years in the same ~1000 participants. (1/9)
new review paper w/
@rtg2753
@yael_niv
and
@LnccBrown
! we discuss multiple mechanisms by which memory is improved/structured by emotionally relevant events, inc. classic ‘affect’ and our more recent work on ‘prediction’ (prediction-error) route💥 (1/6)
🚨 new! albeit paradoxical - predictably good and unpredictable events - in line with dopamine and norepinephrine - BOTH boost RL and (implicit & explicit) memory as a function of learning (n=766).
data + code:
with
@yael_niv
:
1/7
wondering how surprising events change the structure of our memory?
this paper is out! thread below (1/5)
4 experiments + computational model with
@ptoncompmemlab
@yael_niv
@urrational
joy division! large reward prediction error events are better remembered, show intact links with preceding events, YET interrupt the temporal integration of events around them.
4 experiments + computational model with
@ptoncompmemlab
@yael_niv
@urrational
new preprint!
not all surprising rewards (prediction errors) are created equal in memory! (if you're depressed: more surprising negative > positive; if you're not: more surprising positive > negative).
new work with
@yael_niv
original paradigm:
dear hivefMRImind, have you recorded spoken responses in an fMRI experiment? did you use a tool to transcribe those responses? and if so, could you let me know how to save hours of my life? (w/
@yael_niv
,
@OdedBein
)
And found subjects scoring higher on negative factors and PTSD had greater and more negative recall (vice-versa for positive factors). The depression factor and PTSD specifically predicted more non-episodic (or ‘overgeneralized’) memory, a hallmark of impaired memory (4/9)
Memory for March 2020, during the surprising and impactful onset of the pandemic, was consistently more accessible (recalled more and earlier), leading to a large ‘bump’ in autobiographical memory (compare to memory for 2021). These memories were also more negative… (2/9)
Links! (9/9)
Paper:
Supporting info (w/ separate analyses & results):
All data & code:
Preregistration:
Covid-Dynamic study (resources & data available!):
We had 2 competing hypotheses for changes in subjective time in 2020:
#1
pandemic-induced event boundary would dilate (increase) time for events that crossed it or
#2
the lockdown period (with its everyday sameness) would compress (decrease) time. We found evidence for
#2
(6/9)
Analyzing large-scale autobio. memory would not have been possible without recent advances in NLP including work pioneered by
@rubenvgenugten
: automated scoring of autobio. interview, and
@s_michelmann
@ptoncompmemlab
: GPT-3 event segmentation (preprints available!) (8/9)
In fact, the more negative a person’s reported emotional state during a month (collected Apr-Dec 2020), the more likely they were to later retrieve that month and the more negative its sentiment. We then reduced the dimensionality of measures (to 6 factors)… (3/9)
3 years in the making! click for interesting findings and exploration of value-memory mechanisms that while complex read butter-smooth thanks to
@ea_horwath
lucky enough to have had
@sarahdubrow
on this project, there's no substitute for her
Our 1st foray into free recall dynamics/ CMR to understand how reward motivation changes memory organization.
@ea_horwath
knocked it out of the park with
@ninarouhani
. Also one of the few remaining collaborations I had with
@sarahdubrow
.
Overall, an interplay between collective events & individual affect profoundly shaped long-term memory for 2020. This ecological approach is thanks to the methodical & tireless work of coauthors and the Covid-Dynamic Team,
@NeuroDamian
,
@tessa_rusch
& Ralph Adolphs (7/9)
Lastly, new subjects rated memory and sense of time between 2020 collective events (top news items per month). Surprise, positive and negative affect predicted greater memorability, with average surprise reflecting the distribution of autobiography in original sample (5/9)
... inc. the unyielding support of my mentors/labs,
@DShohamy
,
@yael_niv
,
@ptoncompmemlab
, adolphs lab,
@vpmurty
(guru) & my collaborators/friends, boosters of my work and spirit, without whom I would have surely skedaddled. while there is an ‘i’ in science, there shouldn’t be
want to learn how the brain rebounds? absolute slam dunk of a paper and a great neural assist to my behavioral work on prediction error and memory!
(please blame
@JamesWardAntony
for encouraging my persistent punnery)
We are hiring a Professional Specialist for a new NIMH Conte Center for Understanding Latent Cause Inference in Health and Illness shared between Princeton and Rutgers (the job is at Princeton, working directly with me, and is partly a position in my lab). More info:
/1
all my hands are in this cookie jar (and by hands i mean data from my 2018 paper) - congratulations on such thoughtful and rigorous work! i am hungry now.
At long last, my paper with
@cate_hartley
and
@GrassieHannah
on decision making, learning & memory across adolescence is out in
@eLife
. I also made some thematic cookies to help illustrate our study and findings! 1/
we find that large reward prediction error (RPE) events are better remembered, show intact links with preceding events (exp. 1 & 2), yet interrupt the integration of events that came before from those that came after in memory (exp. 3), first assessed by recognition priming (2/5)
@vpmurty
would have been impossible to burn without Deepu tinder!! (not to be confused with the app, but I'm sure at some point in time, that was also on fire 🔥)
in line with a building dopamine RPE at reward cue, which transfers from reward outcome to the cue predicting the outcome with more learning, we find memory for cue events to increase over learning (and only if there is a cue RPE - exp 2 design). (4/7)
we tested these RPE signals on explicit and implicit memory separately, and in both cases found that a signed RPE at cue increased subsequent memory for the reward-predicting event and an unsigned RPE at outcome boosted memory for the reward-outcome event. (6/7)
we next find the same pattern of results when testing temporal memory for events spanning a high-RPE event or not (exp. 4 & replication): intact sequence memory for a high-RPE event and its predecessor, but impaired sequence memory for events across a high RPE (3/5)
@DavidClewett
@vpmurty
@urrational
@ptoncompmemlab
A hypothesis I got to test in a recent paper! 🦝 where we took an ecological and longitudinal approach to studying how collective events and individual affect influence memory (6/6)
Paper:
Tweeprint:
Wonder what 2020 did to your memory or sense of time?
🚨 Now out in
@PNASNews
🚨 we tracked how collective events and individuals’ emotional state (sampled throughout 2020) shaped spontaneous autobiographical memory across 3 years in the same ~1000 participants. (1/9)
Components of the affect route (arousal & motivation) and prediction route (unsigned ‘UPE’ & signed ‘SPE’) may differentially rely on LC-NE versus DA systems to improve memory w/ distinct consequences for memory segmentation versus integration, respectively…. (2/6)
we captured these effects in a variant of the Context Maintenance and Retrieval model (CMR; Polyn, Norman, & Kahana, 2009), modified to incorporate RPEs into the encoding process (5/5)
🚨 new! albeit paradoxical - predictably good and unpredictable events - in line with dopamine and norepinephrine - BOTH boost RL and (implicit & explicit) memory as a function of learning (n=766).
data + code:
with
@yael_niv
:
1/7
we dissociate the effects of signed and unsigned reward prediction errors (RPEs), experienced at the reward-predictive cue and reward outcome, in reinforcement learning (RL) models (where these RPEs update a dynamic learning rate), and in memory for these associated events. (2/7)
consistent with classic models of learning (Mackintosh and Pearce-Hall), which paradoxically posit that attention is enhanced for predictable and unpredictable events, we find that a signed RPE at cue and an unsigned RPE at outcome both update learning and enhance memory. (3/7)
@hawkrobe
@mai_lin_nguyen
@yael_niv
@OdedBein
this is cruelty. every time i've listened to my own voice, i look around at people with a new sense of betrayal "why haven't you told me i need to change my voice and to never laugh spontaneously"
this effect is demonstrated within and between subjects. individually, better learners show greater enhancement of this learning-modulated cue memory. (5/7)
@mariam_s_aly
yes! the 'blursday' database showed social isolation dilated online time, and one could expect that isolation (w/ potentially fewer distinct episodes) would lead to later time compression. would be great to formally test this potential dissociation!
moreover, greater distance between events at encoding led to better sequence memory for events across a low-RPE event, but not a high-RPE event, suggesting separate mechanisms for the temporal ordering of events within versus across a latent reward context (4/5)
in a final choice task, subjects preferred more rewarded and valued cues and outcomes, but when asked to choose between a cue and outcome from a single trial - the signed RPE at outcome 'irrationally' predicted choice, pointing to a hedonic rather than mnemonic effect. (7/7)
@yael_niv
@GregoryRSL
@nathanieldaw
agree with
@nathanieldaw
! would also add that jupyter notebooks (using an R instead of a python kernel) is a nice way to organize R code, and may be useful in teaching modules (i.e., one notebook per module).
@DavidClewett
@vpmurty
We discuss the computational mechanism underlying PE memory boost and segmentation (work w/
@urrational
@ptoncompmemlab
, variant of CMR model
@seanpolyn
), w/ a theoretical illustration of how the pandemic might have served as a high-PE event, boosting memory for that time… (5/6)
@urrational
@paniq
@yael_niv
@ptoncompmemlab
@yeonsoon_shin
@sarahdubrow
I've found that regardless of whether RPEs indicate a new state or not they are better remembered (Rouhani et al, 2018), but this discontinuity/restructuring of memory only occurs when the RPE indicates a change in the latent reward state (as in this paper). (2/2)
@OndrejZika
@yael_niv
that's right (although the parameters used to generate the simulations were fit using the value estimates) - either using values or learning rates checked out though!
@ptoncompmemlab
@yael_niv
@urrational
[1/2] revision! despite joy division's clear prediction of topics important to cognitive neuroscience, the name's association with the Nazis has led us to remove the “before the colon” title.
@OndrejZika
@yael_niv
very important point! we also use trial-by-trial value estimates (learning is 'Pavlovian') and were able to recover the models. here we focused on simulating and fitting trial-by-trial learning rates since these were the only differences between the models (happy to talk more!)
@gershbrain
@ptoncompmemlab
@yael_niv
@urrational
we're on it, also throwing in 24 hours (reward prediction error effects on temporal memory post consolidation) with its computational model, New Dawn Fades (drifting temporal context model account of RPE consolidation effects)
@ptoncompmemlab
@yael_niv
@urrational
[2/2] we also clarify in the discussion that the event boundary occurs across the high reward prediction error (RPE) event, meaning the high RPE event is associated with the events that came before and after it but breaks the link between them.
@paniq
@ManojDoss
context can be a very expansive abstraction for how we integrate or divide our experiences in memory - so the post-LSD experience sounds like a new context to me :)
@QasimEtal
this is great, congrats! a rigorous examination of RPE-memory effects - also in line with the bias we found for the non-depressed ('pos. affect')
interesting to link it to this work too!
@AndreaMCataldo
@DanielGDillon
🎉🎉🎉
@RobynFivush
thank you! given your fascinating work on narrative identity/coherence, college-aged students and the pandemic, i've been meaning to take a closer look at developmental signatures in our data :)
@paniq
interesting! i'm not sure how such synaptic activity may influence long-term potentiation (thus supporting long-term memory), but
@ManojDoss
may know! :)
@SNSmithSNS
@HarvardHBS
❤️ honestly, you absolutely blew me away! (and didn’t need much of me at all ;) looking forward to the great work you’ll be doing!
@DavidClewett
@vpmurty
We next review work showing the effect of signed and unsigned prediction errors, independent from the affective outcome itself, in improving memory across appetitive and aversive domains during reinforcement learning (4/6)