A high fidelity quantification of the time course of learning consolidation from a massive visual search dataset

Le résumé fourni par la source

Visual search behavior adapts to the statistics of local contexts, as shown in cognitive effects like priming (e.g., Maljkovic & Nakayama, 1994), statistical learning (e.g., Geng & Behrmann, 2005), contextual cueing (e.g., Chun & Jiang, 1998), and selection history (e.g., Awh et al., 2012). Previous work has described a general evidence accumulation function that may underlie such learned adaptations in behavior across a range of features (Kramer et al., 2022), but questions remain about the time course of the consolidation of this learning. Here we used a massive dataset (~15.7 million users, ~3.8 billion trials) of human behavioral data from a mobile app (Airport Scanner, Kedlin Co.) to quantify the consolidation of learning in a visual search task over a range of time delays, and to describe the impact of sleep during the consolidation period. The size of this dataset allowed for an extremely high fidelity characterization of the time course of consolidation from seconds to days with a high degree of temporal precision. Linear modeling of the effects of prior experience (proportion of trials containing a target), time delay (amount of time since the last block of search) and their interaction showed significant main effects of prior experience and time delay, as well as a complex nonlinear interaction for all of our dependent variables (hit rate, target present correct response time, correct rejection rate, target absent correct response time). A secondary analysis on the effect of sleep on visual search performance revealed main effects of sleep on search accuracy but no interaction of sleep with accumulation of evidence about target prevalence. These characterizations of the temporal dynamics of consolidation behavior provide much needed constraints on hypotheses about the myriad of neural mechanisms underlying learning and their characteristic time courses.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.