Benutzerhandbuch Datenschutzhinweis Impressum Kontakt





Using the past to estimate sensory uncertainty


Rohe,  T
Department Human Perception, Cognition and Action, Max Planck Institute for Biological Cybernetics, Max Planck Society;
Max Planck Institute for Biological Cybernetics, Max Planck Society;


Stegle,  O
Max Planck Institute for Developmental Biology, Max Planck Society;

Externe Ressourcen

(beliebiger Volltext)

Volltexte (frei zugänglich)
Es sind keine frei zugänglichen Volltexte verfügbar
Ergänzendes Material (frei zugänglich)
Es sind keine frei zugänglichen Ergänzenden Materialien verfügbar

Beierholm, U., Rohe, T., Stegle, O., & Noppeney, U. (2016). Using the past to estimate sensory uncertainty. Poster presented at Computational and Systems Neuroscience Meeting (COSYNE 2016), Salt Lake City, UT, USA.

Combining multiple sources of information requires an estimate of the reliability of each source in order to perform optimal information integration. The human brain is faced with this challenge whenever processing multisensory stimuli, however how the brain estimates the reliability of each source is unclear with most studies assuming that the reliability is directly available. In practice however reliability of an information source requires inference too, and may depend on both current and previous information, a problem that can neatly be placed in a Bayesian framework. We performed three audio-visual spatial localization experiments where we manipulated the uncertainty of the visual stimulus over time. Subjects were presented with simultaneous auditory and visual cues in the horizontal plane and were tasked with locating the auditory cue. Due to the well-known ventriloquist illusion responses were biased towards the visual cue, depending on its reliability. We found that subjects changed their estimate of the visual reliability not only based on the presented visual stimulus, but were also influenced by the history of visual stimuli. The finding implies that the estimated reliability is governed by a learning process, here operating on a timescale on the order of 10 seconds. Using model comparison we found for all three experiments that a hierarchical Bayesian model that assumes a slowly varying reliability is best able to explain the data. Together these results indicate that the subjects’ estimated reliability of stimuli changes dynamically and thus that the brain utilizes the temporal dynamics of the environment by combining current and past estimates of reliability.