Menichini, Elena; (2025) Learning and exploiting sensory statistics: behavioural, computational and neural representations of statistical structures in perceptual decision-making. Doctoral thesis (Ph.D), UCL (University College London).

Text Menichini__thesis.pdf - Accepted Version Access restricted to UCL open access staff until 1 September 2026. Download (28MB)

Abstract

The ability to integrate sensory information with prior knowledge is central to perceptual decision-making. Our goal is to investigate how sensory history and statistical patterns influence real-time decisions. In this study, we explored how humans and rats learn to categorise sounds in a two-alternative forced-choice auditory task, where feedback guides learning. We manipulated stimulus statistics by sampling stimuli from distinct distributions while maintaining equal overall category probabilities. Humans and rats exhibited sensory-dependent biases in their performance, flexibly and gradually adjusting their choice strategies to align with the underlying statistical context, thereby maximising reward consumption to nearoptimal levels. Although sensory-dependent biases were comparable across species, humans and rodents exhibited distinct trial-to-trial learning patterns. We developed computational agents, a boundary-estimation model, which learns to represent the boundary between the two categories, and a stimulus-category model, which learns the distributions of each category independently. The two models recapitulated the sensory-dependent biases and the distinct learning updates observed in experimental data. Human behaviour aligned more closely with the stimulus-category model, whereas rat data displayed a higher degree of individual variability in updating strategies. Furthermore, we uncovered distinct neural signatures in the medial prefrontal cortex (mPFC) and ventral orbitofrontal cortex (vOFC) associated with decision-making and task execution, closely reflecting the behavioural strategies and flexible choice biases exhibited by the subjects. We identified a correlation between context-dependent neural biases in the mPFC and behavioural choice biases, suggesting that the representations of stimulus-to-choice mapping in the mPFC may contribute to flexible and adaptive decision-making. These findings uncover strategies for flexible decision-making, revealing commonalities and differences between species, and advance our understanding of how the brain adapts to implicit changes in sensory statistics.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as CSVJSONXML

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item