Föreläsningar och seminarier Cognitive Neuroscience Club med Valeria Peviani: "Probabilistic computations in body perception"

"Probabilistic computations in body perception"

Föreläsare/talare

Valeria Peviani, Donders Institute for Brain, Cognition & Behaviour, Nederländerna

Abstrakt (på engelska)

Even when we don’t see them, we can feel where our hands are and have a sense of how our fingers are configured in space. This fundamental perceptual ability likely relies on both somatic information, such as proprioceptive and tactile signals, and on a stored representation of body physical features (e.g., phalanx length). Research on hand perception has shown that healthy individuals tend to perceive their fingers as shorter (i.e., fingertips as closer to the knuckles). For more than a decade, this has been taken as evidence that the brain relies on distorted knowledge about hand physical dimensions.

We took a computational approach to formally describe processes behind body perception, quantify and test behavioral predictions, and ultimately shed the light on the origin of these perceptual biases. We found that perceptual biases in body perception may arise from the probabilistic nature of the computations behind body representation, rather than from its distorted content. We hypothesized that, in order to compute the spatial position of the fingertips starting from noisy sensory signals, the brain relies on the Bayesian integration of sensory likelihoods and priors over joint angles and phalanx lengths. Critically, if the integration is performed in transformed (Cartesian) coordinates, priors over angles lead to biases affecting overall finger length, despite unbiased priors over phalanx length. We tested the model by asking twenty participants in VR to indicate the perceived position of their joints and fingertips using a virtual pointer, while keeping the hand at different postures. Using Maximum Likelihood Estimation, we found that perceptual biases, including overall finger length misestimation, can be well described by Bayesian computations occurring in external, Cartesian space. This clearly suggests that perceptual biases in body perception does not necessarily reflect distortions in body representation.