Poster session

Bayesian inference requires the use of numerical solutions since posterior distributions in closed-form are rarely computable in complex models. Popular algorithms and specialized software demand a considerable amount of computational resources and Bayesian analyses requiring hours or days of uninterrupted computation are common. Furthermore, the need for scalable Bayesian methods intensifies as large datasets on diverse domains become readily available. In this work we explore the performance of Consensus Monte Carlo (CMC) in the context of hierarchical models. This distributed algorithm splits the data into several different chunks and assigns each one to a different machine, calculates the posterior distribution corresponding to each data partition, and then mixes them back together to obtain the posterior distribution reflecting the whole dataset, where the final “consensus” distribution is a weighted average of the posterior distributions returned by each machine. We illustrate the workings of CMC by implementing a hierarchical model of choice equilibrium over NFL play-by-play decisions. The dataset includes over a quarter million plays from 2013 to 2023 and, given its moderate size, allows for a direct comparison between CMC and the model implemented in a single machine using all observations at once. The hierarchical model we use as example describes choices between rushing or passing as a function of the relative gain in yards returned by each of those alternatives, and explains deviations from optimal equilibrium in terms of covariates at team, game, and quarter level.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Interrupted time series analysis is a statistical method to study the effects of a deliberate intervention by observing data over a period before and after a change. In this project, we consider interrupted time-series data from a mobile health intervention study aimed at promoting psychological well‐being in college students. We will apply a model for interrupted time series based on the Ornstein-Uhlenbeck (OU) diffusion model, a stochastic time series model whose main parameters capture intraindividual variability, an attractor point or homeostasis level, and an elasticity parameter that governs the speed with which the process returns to its attractor after a perturbation. Interruptions in these time series can be characterized as discrete state shifts in one or more of these parameters, leading to the hierarchical Bayesian interrupted OU model that we apply to the mobile health intervention. We evaluate the intervention's effectiveness by examining the levels of psychological well-being across four study phases: pre-intervention, intervention, immediate post-intervention, and late post-intervention. We operate under the assumption that we can categorize the time series according to these phases, anticipating that participants' psychological well-being tends to stabilize at specific homeostatic levels during each phase. Additionally, we evaluate the applicability of BayesFlow to this broader class of problem. BayesFlow is a new simulation-based inference method that can provide high-efficiency Bayesian parameter estimation even with complex, time-variant models, but its application to a multilevel hierarchical model such as ours requires a thoughtful implementation. We discuss strategies for our specific case and possible expansions of our work.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

The study explores the question: How does schema alter the way people see the world? Specifically, will the brain change its way of processing due to the learning of schema? To investigate this question, we look at how regions of the brain change connectivity after schema learning through Intersubject Correlation (ISC) and Intersubject Functional Correlation (ISFC). The experiment uses naturalistic continuous stimulus. Participants listened to rhyming poems in which each stanza was related to one of ten topics (such as music, religion, animals, or the government). There was a hidden pattern to these stanzas, which was only revealed to participants halfway through the experiment: the topics appeared in a set order, such that (for example) a stanza about music was always followed by a stanza about the government. The results showed that schema learning changed how information flowed in the brain. Before people learned the schema, they had to rely solely on auditory information to understand the poetry. However, after schema learning, people could use their internal knowledge of the sequence order to help understand the poems and rely less on sensory information. This study is important in that it delves deeper into our cognitive processes and enables us to better understand how we combine the past and present to interpret current events.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Investigating working memory (WM) processing development in children is important, as it sheds light on possible educational outcomes. In the current study, we employed spatial and verbal WM games within an online adaptive learning environment to extend the key benchmarks identified by Oberauer and colleagues (2018) to children. Approximately nine thousand children from Dutch schools, aged between approximately six and twelve years old (grades 1 to 5), provided millions of responses and a means for studying WM development in students’ regular school practice environment. The adaptive nature of the employed WM memory games ensured that children responded to items matched to their ability. As such, the games optimally supported learning but also created challenges for analysis of typical lab findings. Nevertheless, our results replicated many classical WM findings observed in adults, including primacy and recency effects, item and order errors, the fill-in effect, and the locality constraint on transposition errors. We used hierarchical Bayesian modeling to assess primacy and recency effects, with varying magnitudes across grade levels and games. Bayesian Chi-Square tests indicated that the order errors (i.e., transpositions) increased with grade level. Additionally, transposition errors clustered around correct serial positions, confirming locality constraints of transpositions. Furthermore, fill-in errors in transpositions occurred more frequently than infill errors in the verbal WM game. In contrast, some item errors (i.e., repetition and intrusion errors) decreased with maturation. Our findings offer insights into the dynamic development of WM processing in children and underscore the reliability of classical WM findings in large-scale adaptive learning environments.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Our perceptual and memory systems constantly make predictions about the near future. Errors in prediction lead to an updating of these systems therefore influencing our subjective experiences of events. In the current study, we evaluated whether prior (event) knowledge and age influence one’s ability to make predictions during complex, dynamic events. In Experiment 1, 140 participants (aged 18-75 years) watched and segmented movies with or without context. In Experiment 2, 41 younger and 28 older adults watched the same movies with the same context manipulation; however, the movie was paused several times, and the participant was asked to make predictions about what would happen 5 sec later. In Experiment 1, we found that context did not influence event segmentation ability. In Experiment 2, we compared young and older adults’ predictions within the same events and across different events as well as predictions when context was and was not provided. Both young and old made more accurate predictions when they had context. No significant age-related differences in prediction accuracy were observed. Although Older adults made better predictions in the same event, and Younger adults made better predictions across events.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Individuals’ thoughts often wander during sustained attention tasks, particularly when the task at hand is understimulating and monotonous (Smallwood & Schooler, 2015). Often, these episodes are directed toward segments of music that are recalled involuntarily, sometimes called “earworms” (Halpern & Bartlett, 2011). While earworms are a normal phenomenon in music cognition, their potential to disrupt concurrent task performance is less understood, particularly when they compete for attentional resources. Here, we introduce a computational framework for examining the effects of involuntary musical imagery (INMI) on a continuous working memory performance task. Specifically, we use a comprehensive model of performance in the N-back task to predict responses when attention shifts between the extant task (“on-task”) and an earworm (“off-task”). Model predictions are compared against data collected from 203 participants who self-reported INMI prevalence after performing 1-, 2-, or 3-back tasks. We show that response errors are linked to increases in self-reported INMI, where periods of off-task behavior dominate attentional resources and prevent the refreshing of target items in working memory.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

We compare storytelling in GPT-3.5, a recent large language model, with human storytelling. Although GPT models are capable of solving novel and challenging tasks and matching human-level performance, it is not well understood if GPT processes information similarly as humans. We hypothesized that GPT differs from humans in the kind of memories it possesses, and thus could perform differently on tasks influenced by memory, such as storytelling. Storytelling is an important task for comparison as GPT becomes an increasingly popular writing and narrative tool. We used an existing dataset of human stories, either recalled or imagined (Sap et al., 2022), and generated GPT stories with prompts designed to align with human instructions. We found that GPT's stories followed a common narrative flow of the story prompt (analogous to semantic memory in humans) more than details occurring in the specific context of the event (analogous to episodic memory in humans). Furthermore, despite lacking episodic details, GPT-generated stories exhibited language with greater word affect (valence, arousal, and dominance). When provided with examples of human stories (through few-shot prompting), GPT was able to align its stories' narrative flow with human imagined stories but not human recalled stories. GPT was unable to match its affective aspects with either human imagined or recalled stories. We discuss these results in relation to GPT's training data as well as the way it was trained.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

The Theory of Visual Attention (TVA; Bundesen, 1990) is a theoretical framework and robust computational model, which explains and predicts how humans select and process visual information. We introduce RStanTVA, a novel implementation of TVA for partial and whole report, programmed in Stan and R. The software addresses several limitations of predecessors, such as platform dependence, licensing, and inability to fit Bayesian and hierarchical models. RStanTVA increases the flexibility and accessibility of TVA modeling by streamlining model generation, fitting, and analysis. We replicate a range of previously published experimental findings, and demonstrate the advantage of Bayesian parameter inference for TVA models.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Contrary to common intuition, a group of people recalling information together remembers less than the same number of individuals recalling alone (i.e., the collaborative inhibition effect). To understand this effect in a free recall task, we build a computational model of collaborative recall in groups, extended from the Context Maintenance and Retrieval (CMR) model, which captures how individuals recall information alone. We propose that in collaborative recall, one not only uses their previous recall as an internal retrieval cue, but one also listens to someone else's recall and uses it as an external retrieval cue. Attending to this cue updates the listener's context to be more similar to the context of someone else's recall. Over an existing dataset of individual and collaborative recall in small and large groups, we show that our model successfully captures the difference in memory performance between individual recall and collaborative recall across different group sizes from 2 to 16, as well as additional recall patterns such as recency effects and semantic clustering effects. Our model further shows that collaborating individuals reach similar areas in the context space, whereby their contexts converge more than the contexts of individuals recalling alone. This convergence constrains their ability to search memories effectively and is negatively associated with recall performance. We discuss the contributions of our modeling results in relation to previous accounts of the collaborative inhibition effect.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

In three preregistered experiments (Ntotal = 1127), we investigated the role of source memory in the processing of advisor expertise and accuracy in advice taking. Participants received advice about the job as a doctor from two different sources. In Experiment 1 (N = 619), advice was provided by a doctor (i.e., an expert) and a lawyer (i.e., a lay person). In Experiment 2 (N = 248), advice was provided by a labeled high- or low-performing advisor (i.e., a valid or invalid source, respectively). Experiment 3 (N = 260) replicated the procedure of Experiment 2 with more distinct source labels. After the advice presentation, participants provided their own estimates and completed a source monitoring task. We applied multinomial processing tree modeling to measure source memory independent of guessing and combined this method with a regression approach to measure its moderating effect on advice weighting. In all experiments, the influence of advisor quality was stronger the better the memory for the source, suggesting that the actual weighting occurs at the time of judgment. The presented research presents a methodological extension of the classic advice taking paradigm and establishes source memory as an important constraint for adaptive advice weighting. The multinomial modeling approach offers a nuanced understanding of the role of advice sources and memory in advice taking. Overall, the presented research highlights the advantages of statistical modeling applications in applied memory contexts.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Episodic memory and schema knowledge are known to interact when we recall everyday events – such as the location of an object in a scene. It may seem intuitive that when our schema knowledge does not match a given scene (e.g., the object location in the scene is incongruent with our schema), recall would rely more on episodic memory. However, the process that controls this tradeoff is unclear. For example, Ramey and colleagues (2022) conducted an experiment using a spatial recall task with natural scenes using the Recollection/Familiarity paradigm. They concluded that episodic memory controls the amount of prior knowledge used in recall. They argued that strong episodic memories suppress schema knowledge during retrieval. They found greater accuracy for congruent versus incongruent scenes, with greatest accuracy for new scenes, decreasing for familiar scenes, and eliminated for recollected scenes. We directly replicated Ramey’s experiment, and found that the general effect is robust. However, we take a different view on this competitive relationship where memory suppresses schema knowledge, namely that there is a trade-off between the strength of episodic memory and schema strength. We implemented a Hierarchical Bayesian model that assumes a relative weighting of schema knowledge versus episodic memory that depends on the strength of the episodic memory. We model familiarity/recollection as memory strength, and the congruent versus incongruent conditions as having different priors - with the congruent prior linked to accuracy for congruent new scenes in the experimental data, and the incongruent prior linked to accuracy for incongruent new scenes. Our model simulations are consistent with the experimental results without assuming that episodic memory controls schema knowledge, but instead allows for a bi-directional influence.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

People frequently encounter event (going shopping) and location (market) cues in their everyday environments. Sometimes these cues prompt people to think of a previous event or simulate a future event. In Sheldon and Chu (2017, QJEP), participants generated a greater number of memories and rated them as more vivid for event than for location cues. The current study extends those findings to investigate how future event simulations are influenced by cue type, and how they might compare with simulations of past events. During an event fluency task, 40 participants were shown 12 event or location cues and were instructed to generate as many cue-related future and past events as they could. Participants produced more future and past event simulations for event than for location cues. We used novel chained connections analyses to gain further insight into how people simulate multiple events. This analysis investigated the unfolding of chained-event sequences consisting of multiple, related events. For each chain that a participant generated, they were shown five types of connections (activities, place, people, time, and objects) and were asked to identify the one that connected each pair of adjacent events. For future and past chains, event-cued chains were typically connected by activities and people whereas location-cued chains typically were connected by locations and activities. The similar pattern of results for future and past events supports that they are simulated through similar mechanisms, and that the initiation and unfolding of these simulations is influenced by the information present in a person's environment.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

In open-ended decisions, options are often ill-defined and must be generated by the decision maker. How do people make decisions without predefined options? Our study explored this question using 30 consumer products under both time-free and time-pressure conditions. We found that while people prioritize early-generated options under time pressure, their decision quality remains as good as in time-free conditions. Additionally, computational modeling showed that decision makers decide while generating the options from memory. Together, our behavioral and model-based findings shed light on the cognitive mechanisms of memory-based decisions, advancing the understanding of open-ended decisions.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

The sense of agency (SoA) is the experience of control over actions and outcomes. Though SoA is a fundamental human experience, there is no consensus on the mechanism or theoretical explanation behind it. Additionally, there are few formal models. One proposed explanation of SoA is that it can be explained as a function of an individual's goals: if something (internal or external) prevents the individual from making progress on their goals, their sense of agency should decrease (Saad et al., 2024). We extend this explanation and posit that a participant’s SoA is the result of a continuous process by which the participant is evaluating evidence about their actions and outcomes in reference to their goal. To test this explanation, we apply a classic evidence-accumulation modeling framework, the drift diffusion model (Ratcliff, 1978) to data from two online behavioral experiments where participants’ sense of agency is manipulated. Neither bias (beta = 0.8) nor boundary separation (alpha = 6) varied across conditions or experiments. To capture the differences in evidence accumulation across conditions, we propose a two-stage mechanism that determines the rate of evidence accumulation (i.e., drift rate) as a function of the current distance to goal and an expectation evaluation. We use task time as a behavioral proxy for distance to goal. Model simulations suggest that when the experimental manipulations prevented the participants from reaching their goal, that both the drift rate and the participant’s SoA decreased. We address implications, limitations, and directions for future work.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

In our daily lives, we routinely search for items we have placed in various locations (e.g., car keys, cellphones). Often, the items we seek are ones that we have placed in these locations ourselves! We investigated whether contributions of such self-generated movement improve the precision of location recall. Furthermore, we explored embodied contributions to object-location memory by investigating how implicit motor simulation, constitutive to the identification of objects from specific categories, may interact with enactment. In a modified version of Tompary et al.'s (2020) recall paradigm, participants learned associations between images (animals or tools) and their locations along the perimeter of a circle. During study, participants learned these locations either by actively placing items in target locations or by passively observing as items moved to these locations. At test, participants were asked to place each item in its learned location. We offer evidence (N = 18, collection ongoing) that enactment—moving an item during learning—leads to more precise recall of object-location associations. Enactment also induced a boost in memory confidence and reduced response times. Results suggest that motor information related to tool-category membership contributes to the precision of object-location recall; however, this effect does not appear to interact with enactment. Analyses of mouse-tracking trajectories suggest additional differences in retrieval processes based on encoding conditions. This research represents a first step towards using more naturalistic experimental paradigms for studying memory for object locations with a set-up that captures key aspects of things that people do every day.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Previous research suggests that navigation involves weighted integration of various sensory inputs such as visual and body-based cues. When we navigate an environment, navigational goals (such as travelling a certain distance or turning a specific angle) are often involved, which leads to internal predictions about the navigational outcomes. To examine the role of internal predictions in the cue integration process, our study focuses on how humans integrate these multi-sensory cues to navigate with a goal. To study how prediction errors are resolved in naturalistic navigation, we used immersive Virtual Reality (VR) to replicate real-world navigation scenarios. Participants wore a wireless VR head-mounted display that enabled free movement and naturalistic navigation in a virtual room with landmarks. Using VR, we introduced a mismatch between visual and body-based cues such as shifting the landmarks in the virtual room. Participants rotated in place with visual landmarks to a certain angle and then rotated back in darkness as a response. A quick flash of the shifted landmarks was shown during the responding phase. The trajectories of the learning and responding rotations were recorded. We found that participants adapted their rotational velocities not only in response to external cues (landmarks) but also based on internal goals (goal angles). Specifically, when visual and body-based cues were misaligned, participants adjusted their velocities depending on the shifted landmark, suggesting an integration process between perceived external information (landmarks) and internal representations (expectations). Building on this empirical evidence, we proposed a closed-loop feedback controller model, based on Linear-Quadratic-Gaussian control, to predict trajectories over time. This model allows us to test the hypothesis on metabolic energy optimization during rotation through controlling velocity. By fitting parameterized feedback gain and observer (Kalman) gain to trajectories from individual trials, the model is able to predict turning trajectories relatively well. Across all participants, the eigenvalues of the systems were found to be within the unit circle, confirming the stability of the model. Our findings support Perceptual Control Theory, suggesting that navigation is not merely a causal response to external stimuli. Instead, it is part of a closed-loop system designed to stabilize internal representations against external disturbances. In this model, response to stimuli serves as feedback to modify the model inputs which are our representations of the environment and ourselves. This feedback control ensures that our representations are continuously updated to align with our goals. In sum, our findings shed light on how we minimize the discrepancies between perceived inputs and desired states to navigate effectively.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Judgments about objects in the world are often based on probabilistic information (or cues). A frugal judgment strategy that utilizes memory (i.e., the ability to discriminate between known and unknown objects) as a cue for inference is the recognition heuristic (RH). The usefulness of the RH depends on the structure of the environment, particularly the predictive power (validity) of recognition. Little is known about the trajectories in use of the RH across the lifespan. This project provides an overview of the extent to which primary schoolchildren (M = 9 years), older adolescents (M = 17 years), younger adults (M = 25 years), and older adults (M = 71 years) recruit the RH when making judgments and around what age adaptive use of the RH emerges. All participants made comparative judgments in task environments with either high or low recognition validity. Reliance on the RH and validity of recognition memory was measured with a Bayesian hierarchical multinomial modeling approach. Results indicated that already schoolchildren make systematic use of the RH but that adaptive strategy selection emerges considerably later. These findings suggest that the use of simple memory heuristics does not progress unidirectionally across development but strongly depends on the task environment, in line with the perspective of ecological rationality. Adaptive heuristic inference seems to require experience and a developed knowledge base.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Pilots rely visual information from either the runway or from indicators on the primary flight display to successfully land an aircraft on a runway, which can be compromised by unexpected events (e.g. laser pointers aimed at the plane from the ground). These disruptions can have devastating results if occurring during a landing sequence, particularly towards the end of the landing sequence. The current practice is to abort the landing sequence followed by another attempt; however, subsequent attempt(s) can be costly and are not guaranteed success. The long-term goal of the current research study is to incorporate auditory guidance as part of a multimodal navigation system that would complement visual displays to compensate for visual disruptions during a landing sequence. We previously pilot tested potential auditory signals to incorporate into the multimodal navigation system and found up to 6% mean error rate when prompted to report direction of intended navigation as indicated by auditory signals. Due to the limitations of mean accuracy, our current proposed study will use General Recognition Theory (GRT), a more advanced perceptual model, to further assess perception of the auditory signals when paired with visual display in two separate experiments (Experiment 1: vertical guidance; Experiment 2: horizontal guidance). For vertical guidance, auditory signals will either present two tones sequential or one tone with pitch modulation (between-subject) to indicate upward or downward adjustment. For horizontal guidance, left and right headphone presentation will indicate left and right direction adjustments. Both experiments will use the same visual display currently used by pilots for vertical and horizontal guidance. Each experiment will consist of four auditory/visual stimuli (Exp. 1: up/up, up/down, down/up, down/down; Exp. 2: left/left, left/right, right/left, right/right) with four response options of perceived auditory/visual combination. We anticipate that congruent auditory/visual signals (Exp. 1: up/up, down/down; Exp. 2: left/left, right/right) will violate perceptual independence, while incongruent signals (Exp. 1: up/down, down/up; Exp. 2: left/right, right/left) would not. A violation of perceptual independence for congruent but not incongruent signals would indicate a unified perception of adjustment direction. The strength of perceptual dependence will be used to evaluate the effectiveness of each of the two tone presentations.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Investigations were made as to how novelty impacts autobiographical event memorability via daily diary collection. Participants described three events that took place each day for two weeks, generating a unique title for each event. After a two-week delay, participants were presented with a subset of those titles and asked to rate their memory vividness. Events reported as being ‘new’ (never having occurred before) were remembered with the greatest vividness, followed by ‘periodic’ (occurs occasionally), then ‘routine’ (occurs every day) events. A subsequent question was whether the relative semantic similarity (RSS) of an autobiographical event modulates this effect. Through the use of SBERT models, the semantic similarity between one event and a participant’s remaining events were quantified, outputting a series of pairwise cosine similarity values. These values were averaged across events to create an event-level, RSS variable. We found that routine events with low RSS were remembered with greater vividness than routine events with high RSS. However, the inverse of this relationship was found in new events. This suggests that novelty supports memory vividness, but does so more effectively when there is an underlying scaffold of familiarity. Further, routine events, which were the least vivid, demonstrated a vividness boost when they are more semantically ‘unique’. These findings suggest that novelty and schemas may interact to support autobiographical memory retrieval. Further, utilizing SBERT to analyze narrative data has only become a recent possibility, and has the potential to be an exciting new tool given its efficiency and flexibility.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).

Our everyday lives are comprised of multidimensional events involving people, places, and emotions, both new and familiar. A core focus of memory research is to understand how these features may influence memory. While much prior work has investigated these facets of memory using controlled laboratory studies, the current study aimed to understand how features of real-world experiences influence autobiographical memory. To this end, we enrolled participants in an intensive longitudinal “daily diary” study that asked participants to record a wide range of rich information about their experiences each day for two weeks. Participants reported written descriptions of three events that they had engaged in each day, as well as quantitative metrics of novelty of these experiences and their day in general (e.g., how typical a day felt, whether they visited a new location). The written event descriptions were used to prospectively test participants’ autobiographical memory after a two-week delay. Our findings suggest that novelty bolsters both subjective vividness and objective level of detail reported in the memory test. Furthermore, we find that the benefit of novelty extends to other non-novel events that occurred within the same day, and that multiple sources of novelty independently and cumulatively improve memory. These data suggest that novel experiences enhance memory, and that daily diaries are a valuable method to naturalistically investigate these processes.

This is an in-person presentation on November 21, 2024 (13:00 ~ 14:10 EST).