Using the past to estimate sensory uncertainty.

Area of Science:

• Neuroscience
• Cognitive Science
• Psychophysics

Background:

• Accurate environmental perception relies on the brain's estimation of sensory uncertainty.
• Existing perceptual inference models typically assume sensory uncertainty is computed instantaneously and independently for each stimulus.

Purpose of the Study:

• To investigate whether the brain's sensory uncertainty estimates are computed instantaneously or integrate temporal information.
• To challenge the assumption that sensory uncertainty depends solely on the current stimulus.

Main Methods:

• Conducted four psychophysical experiments involving human observers localizing auditory signals with synchronous, spatially disparate visual signals.
• Manipulated visual noise dynamically over time, using continuous changes and intermittent jumps.
• Analyzed audiovisual integration weighted by sensory uncertainty estimates.

Main Results:

• Observers integrated audiovisual inputs, weighting them by sensory uncertainty estimates that combined past and current signal information.
• This integration process aligns with an optimal Bayesian learner, approximated by exponential discounting.
• Sensory uncertainty estimates were not solely dependent on the current stimulus.

Conclusions:

• The brain actively utilizes the temporal dynamics of the external environment for perceptual inference.
• Sensory uncertainty estimation involves combining prior experiences with new incoming sensory data, rather than instantaneous computation.
• Findings challenge current models of perceptual inference, highlighting the importance of temporal integration in Bayesian learning.