Abstract:
Making the optimal decision in the face of multiple sensory inputs from different modalities is essential for survival in a dynamic environment. A key question is how the brain combines sensory cues when they do not agree and there are nonstationary contingencies between sensory cues and action outcomes. We address this question in a combined electrophysiological, behavioural and computational study using a visuotactile two-choice orientation categorisation task in rodents. Rats were trained to categorise the orientation of a 10cm-diameter disc with alternating black and white ridges and grooves by licking at one of the two reward spouts. On every trial, rats explored the disc by vision (V), or vibrissal touch (T), or both (VT). In the T condition, rats categorised orientations 0-25' as horizontal, and 25-90' as vertical. In V condition, however, the category boundary was shifted to 65'. For VT trials, either boundary was applied randomly with equal probabilities. Extracellular recordings were performed from posterior-parietal and frontal cortices. Psychometric curves (choice likelihood vs orientation) revealed a performance enhancement for VT with congruent cues compared to V and T. A cascade model with non-sensory components (lapses and bias) and an integrator of sensory modalities minimising the variance of perceptual noise well-fitted the data. We further observed a cross-orientation Bayesian judgement, as an evidence of generalisation to certain ques. We propose a Bayesian framework to formulate the behaviour of rats in multisensory categorisation with discrepant cues. An economical reinforcement model based on this framework explains the results, and further makes predictions about other behavioural scenarios

Venue:
School of Cognitive Sciences, Institute for Research in Fundamental Sciences (IPM),
Opposite the ARAJ, Artesh Highway, Tehran, Iran on map




* To join the mailing list:
Send an empty email to: scs at ipm dot ir on the subject of: #join