Active Inference, tracking eye movements and oculomotor delays

Tracking eye movements face a difficult task: they have to be fast while they suffer inevitable delays. If we focus on area MT of humans for instance as it is crucial for detecting the motion of visual objects, sensory information coming to this area is already lagging some 35 milliseconds behind operational time – that is, it reflects some past information. Still the fastest action that may be done there is only able to reach the effector muscles of the eyes some 40 milliseconds later – that is, in the future. The tracking eye movement system is however able to respond swiftly and even to anticipate repetitive movements (e.g. Barnes et al, 2000 – refs in manuscript). In that case, it means that information in a cortical area is both predicted from the past sensory information but also anticipated to give an optimal response in the future. Even if numerous models have been described to model different mechanisms to account for delays, no theoretical approach has tackled the whole problem explicitly. In several areas of vision research, authors have proposed models at different levels of abstractions from biomechanical models, to neurobiological implementations (e.g. Robinson, 1986) or Bayesian models. This study is both novel and important because – using a neurobiologically plausible hierarchical Bayesian model – it demonstrates that using generalized coordinates to finesse the prediction of a target's motion, the model can reproduce characteristic properties of tracking eye movements in the presence of delays. Crucially, the different refinements to the model that we propose – pursuit initiation, smooth pursuit eye movements, and anticipatory response – are consistent with the different types of tracking eye movements that may be observed experimentally.

width=50%,align="right"

reference


All material (c) L. Perrinet. Please check the copyright notice.


This work was supported by European Union project Number FP7-269921, "BrainScales".
BrainScaleS logoFET logoFP7 logoEU logo


TagPublicationsArticles TagYear14 TagFreeMove TagBrainScales

welcome: please sign in