Search Papers | Poster Sessions | All Posters
Contributed Talks IV
Talk Session: Friday, August 9, 2024, 10:00 – 11:00 am, Kresge Hall
10:00 am
Excessive neural replay of aversive and uncertain options predicts human irrational decision-making
Tricia Seow1 (), Jessica McFadyen1, Raymond Dolan1,2, Tobias Hauser1,3; 1University College London, 2Beijing Normal University, 3University of Tübingen
Choice deliberation is guided by the uncertainties of available options and their associated outcomes. However, it is unclear how these choice components are involved in the brain’s decision process. Neural replay, a neuromechanism involving the rapid sequential reactivation of states, has recently been proposed to underlie human cognition including value-based decision-making, but yet is unclear how outcome value and uncertainty are involved. With magnetoencephalography (MEG) recordings during a gambling-style task (N=30), we probed the role of replay in evaluating outcome value and uncertainty for choice. We found that forward replay increased for option paths with more aversive outcomes and greater uncertainty during deliberation, which then predicted irrational choices. Moreover, we observed that individual differences in obsessive-compulsive tendencies exacerbated the modulation of value and irrational choice related replay. These findings highlight the significance of replay dynamics in prospective deliberation involving value and uncertainty, and suggest a mechanistic explanation for how these processes may go awry in psychopathology.
10:12 am
Paradoxical replay maintains unbiased and robust representations of task structure
Hung-Tu Chen1 (), Matthijs van der Meer1; 1Dartmouth College
Experience replay is a powerful mechanism to learn efficiently from limited experience. Despite several decades of striking experimental results, the factors that determine which experiences are selected for replay remain unclear. A particular challenge for current theories is that on tasks that feature unbalanced experience, rats paradoxically replay the less-experienced trajectory. To understand why, we simulated a feedforward neural network with two regimes: rich learning (structured representations tailored to task demands) and lazy learning (unstructured, task-agnostic representations). We find that rich, but not lazy, representations degrade following unbalanced experience, an effect that could be reversed with paradoxical replay. To test if this computational principle can account for the experimental data, we examined the relationship between paradoxical replay and learned task representations in the hippocampus. Strikingly, we find a strong association between the richness of learned task representations and the paradoxicality of replay. Taken together, these results suggest that paradoxical replay specifically serves to protect rich representations from the the destructive effects of unbalanced experience.
10:24 am
Multi-goal spatial navigation is mediated by predictive representations with episodic replay in the human brain
Christoffer J. Gahnstrom1 (), Russell A. Epstein1; 1University of Pennsylvania
What are the neural and computational mechanisms underlying human spatial navigation? Previous studies have suggested that reward prediction and replay might underlie key navigational components such as credit assignment, memory consolidation, and planning. However, these mechanisms are usually tested with relatively simple paradigms, making it is unclear what role they might play in ecologically realistic navigational tasks involving rapidly changing goal locations. To investigate this issue, we scanned participants (N=15) with fMRI while they performed a “taxi-cab” task in a virtual city with multiple possible goals. We found that a successor representation model incorporating episodic replay (SR-DYNA) best fit the observed human behavior. To identify the possible neural systems underlying SR-DYNA, we analyzed BOLD activity in terms of several components of the model. We observed parametric tracking of successor state values in anterior hippocampus, parametric tracking of successor prediction error in a network of cortical regions previously implicated in visuospatial memory, and evidence for remote context-dependent episodic replay in the posterior hippocampus. Our results provide behavioral and neural evidence for predictive representations imbued with episodic reactivations as a plausible mechanism of human flexible navigation.
10:36 am
Learning and use of reward-related representations across cortex and time
Ai Phuong Tong1 (), Vishnu Sreekumar2, Kareem Zaghloul2, Mark Woolrich1; 1University of Oxford, 2National Institutes of Health
Reward-related representations are found distributed throughout many human subcortical and neocortical regions that support different neural processes. These representations get used at different points in time for related tasks. However, the way these representations get re-used and strengthen over time is not well understood. To investigate this, we recorded from the temporal lobe and prefrontal cortex with intracranial electrocorticography (ECoG) while human subjects learnt two-choice decisions between two scenes. Subjects were able to straightforwardly re-use knowledge only when reward contingencies stayed the same between the two scenes. Using a Bayesian learner, we inferred reward expectations from choice behavior, and then measured representations of reward expectation in ECoG data. Reward expectations were uniquely represented in distributed regions across human cortex. The representations of reward expectation in the medial temporal lobe and orbitofrontal cortex were re-used between the two scenes, only when subjects could straightforwardly transfer knowledge between the two scenes. Finally, in a separate region, the anterior temporal lobe, the strength of reward representations as measured by similarity between scenes, increased as learning increased. Our findings suggest that patterns of activity representing reward information are integrated into multiple brain regions, get re-used in similar situations, and increase in fidelity with learning.
10:48 am
Sleep Inspires Insight: a Preregistered Study
Anika Löwe1 (), Marit Petzka1, Maria Tzegka1, Nicolas Schuck1; 1UHH
Humans sometimes have insights, which is expressed in a sudden and drastic performance improvement on the task they are working on. While the origins of insights are unknown, previous work has suggested that insights require a form of memory restructuring that often occurs during sleep. In addition, computational work has suggested that neural variability could increase the likelihood of an insight. Although previous work has investigated sleep as a potentially enhancing factor of insights, the evidence for this idea has so far been mixed. One reason for this unclear picture could be that different sleep stages have differential effects on insights. To investigate the link of different sleep stages and variability to insight, we conducted a preregistered study in which N = 90 participants performed an insight task before and after a 20 minute daytime nap. We find that N2 sleep, but not N1 sleep increases the likelihood of insight moments after the nap, suggesting the need for deeper sleep in order to gain insight. Further, analyses of EEG power spectra showed that 1/f slopes could predict insight above and beyond sleep stages. Our findings thus point towards a role of N2 sleep and aperiodic, but not oscillatory, neural activity for insight.