Abstract
Predictive information plays a major role in the control of eye movements. When a visual event can be predicted with some confidence the delay to initiate an oculomotor response is reduced and anticipatory movements oriented toward the predicted event can be observed. These effects of predictability unveil the expectancy state (or prior) of the visuomotor system. Here we try to infer some general properties of the internal representation of a visuomotor prior and its trial-by-trial buildup, by parametrically manipulating uncertainty (thus predictability) in a visual tracking task. We analyze anticipatory smooth pursuit eye movements (aSPEM) in human subjects, when the relative probability p of occurrence of one target motion type (Right vs Left or Fast vs Slow target motion, in two experiments) was varied across experimental blocks. We observed that aSPEM velocity varies consistently both as a function of the recent trial-history (local effect) and as a function of the block probability bias p (global effect). A single model based on a finite-memory, Bayesian integrator of evidence allows to mimic both local and global effects. The comparison of model predictions (through numerical simulations) and data suggest that: aSPEM are based on an internal continuous estimate of the probability bias p (as reflected by the unimodal distribution of aSPEM) the estimate of p is updated according to an (almost) optimal model of integration of probabilistic knowledge, accomodating experience-related and newly incoming information (current trial). This integration leads in particular to asymptotic linear dependence of mean aSPEM upon p and aSPEM-variance proportional to p(1-p) an additional gaussian motor noise with variance proportional to the square anticipatory velocity affects aSPEM. We conclude that the analysis of anticipatory eye movements may open a window on the dynamic representation of the Bayesian Prior for simple visuomotor decisions.
FACETS IST/FET 6th Framework.