Memory & Perception

In this work, we propose a neural network model for free recall that draws direct parallels between neural machine translation (NMT) and cognitive models of memory search, specifically the Context Maintenance and Retrieval (CMR) model. We hypothesize that NMT advancements such as attention mechanisms (Luong et al., 2015) closely resemble how humans reactivate prior contexts (“mental time travel”; Tulving, 1985). To demonstrate these parallels, we train a seq2seq model as a cognitive model of memory search, and evaluate behavior against available human free recall data. We find that at intermediate levels of training, the model can capture several phenomena observed in human free recall experiments (Kahana et al., 2022); and after optimization, the model demonstrates the same optimal behavior as previously derived by the CMR model (Zhang, Griffiths, & Norman, 2023). Performing an ablation study, we demonstrate that behavioral differences between models with and without attention aligns with impaired behavior observed in hippocampal amnesia patients (Palombo, Di Lascio, Howard, & Verfaellie, 2019).

This is an in-person presentation on July 22, 2024 (15:20 ~ 15:40 CEST).

Recent theories of decision making have explained this behaviour using mental sampling mechanisms where people imagine possible outcomes from the available options to guide their choices. Simultaneously, work in other domains has found evidence of particular mental sampling patterns, such as autocorrelations between samples and moderation by prior assumptions, which current decision making theories do not generally consider. Here, we seek to unify this work, developing a new sampling model of preferential choice incorporating these findings in other domains. Our model, based on the Autocorrelated Bayesian Sampler, predicts choice, reaction time, confidence and valuation from a common underlying sampling process. We find a strong correspondence between our model’s predictions and empirical choice data, though performance remains below leading explanations for such tasks. Our model does however cover a broader set of response types than existing theories, suggesting the advantages of considering of a wider range of behaviours than are commonly examined in current decision making studies.

This is an in-person presentation on July 22, 2024 (15:40 ~ 16:00 CEST).

Recent attempts have been made to integrate realistic orthographic representations from the reading literature in global matching models of memory where all orthographic features, namely letters or bigrams were weighted equally. The current study aimed to further extend this equal weight model and explore the consequence of integrating orthographic feature frequency in global matching models of recognition through two approaches. First, we conducted two recognition memory studies where letter or bigram frequency was varied within participants. Results showed feature frequency effects where words comprised of rare letters and bigrams showed higher HR and lower FAR than words comprised of more common letters and bigrams. Orthographic representations (slot coding and bigram models) were situated as individual global matching models where letters/bigrams were either weighted equally or as a linear function of their natural frequency values, and fitted to individual participants' data in a Hierarchical Bayesian framework. Results consistently favoured the feature frequency models over the equal weight models. Second, equal weight and feature frequency global matching models were fitted to a large recognition memory dataset on the level of individual trials to compare their ability in capturing item-level memorability variations across individual words. Model selection results supported the feature frequency model and showed that weighting features by their distinctiveness improved the models ability to capture variability in hit rates across individual words.

This is an in-person presentation on July 22, 2024 (16:00 ~ 16:20 CEST).

Locating incoming sensory information is an essential element of sensorimotor transformations. In tactile localisation, conflicting information from different reference frames can cause judgment errors and slow processing. For example, people are faster and more accurate in locating stimuli delivered to uncrossed compared to crossed arms, presumably because the crossed posture involves conflicting information between the anatomical ("which side of my body") and external ("which side of the room") spatial information. The difference in performance between crossed and uncrossed conditions is known as the ‘crossing effect’. In the current study, we examined to what extent tasks with crossing effects are suited for studying individual differences, focusing on: 1) the within-subject reliability of crossing effects for individual tasks, and 2) the generalisation of crossing effects across different tasks. Ninety-one participants completed three tactile perception tasks that involved uncrossed and crossed limb postures. Two weeks later, they returned to the lab to complete the same three tasks again. Two of the three tasks exhibited good within-subject reliability for both reaction time and accuracy, as evidenced by high intra-class correlations. However, crossing effects did not correlate between different tasks, even though processing speed did. These findings indicate that cognitive processes may generalise poorly across tasks even when they show within-subject reliability over time.

This is an in-person presentation on July 22, 2024 (16:20 ~ 16:40 CEST).

Bistable perception is a fascinating psychological phenomenon in which the conscious perception that a person experiences for a single image spontaneously changes across time. In our experiment, twelve participants observed a dynamic display of moving dots, which initially appeared to rotate in one direction before seemingly reversing . The study consisted of six sessions, during which participants observed sequences of these moving dots across five one-minute blocks. They were instructed to report any perceived changes in the rotation’s direction – specifying shifts to the left, to the right, or expressing uncertainty. Each report of perceived directional change, along with the time taken to notice such changes, was recorded. This process yielded a detailed dataset of response patterns and the timing of perceptual switches. We developed two models to account for the results: a quantum walk versus a Markov random walk model, both with reflecting bounds and measurement operators designed to detect the three responses. Both models required fitting 6 parameters to each session (5 minutes of viewing) using maximum likelihood methods, and then we computed a G^2 lack of fit measure for each model, person, and session. The results, averaged across the 6 sessions for each participant, indicated that the quantum model produced a better fit for 9 out of the 12 participants as compared to the Markov model; however, the advantage was small.

This is an in-person presentation on July 22, 2024 (16:40 ~ 17:00 CEST).