|
|
|
| e-Pub |
Previous | 
Home
Section: New Results
Visual Neuroscience
Bifurcation Study of a Neural Fields Competition Model with an Application to Perceptual Switching in Motion Integration
Participants : James Rankin, Andrew Meso [Institut de Neurosciences de la Timone, UMR 6193, CNRS, Marseille, France] , Guillaume S. Masson [Institut de Neurosciences de la Timone, UMR 6193, CNRS, Marseille, France] , Olivier Faugeras, Pierre Kornprobst.
In this work we have investigated the underlying mechanisms that gate multistable perception, by focusing on the presentation of 1:1 barber pole during long presentations, which is perceived to move in a direction that changes every few seconds. This phenomenon has been studied from the perspective of dynamical systems modeling and human psychophysics: From a modeling point of view, numerical tools from bifurcations analysis were applied to the study of a competition model posed as a feature-only neural field equation (with a continuous feature space) where adaptation and noise are implemented as mechanisms that can drive activity switches. Human psychophysics experiments were jointly done by INT (Institut de Neurosciences de la Timone, Marseille): Human observers were presented a moving grating stimulus over 15s while eye movements and reports of perceptual switches were recorded. Investigating the stimulus contrast, we found that the peak in switching rate observed experimentally occurs close to a bifurcation in the model that separates two mechanistic regimes. By identifying signatures of the switching predicted by the model with the behavioural data at different parts of the transition-contrast curve, we found for the first time, evidence for a dominance of driving mechanisms which shifts from noise dominated at low contrasts to adaptation dominated at higher contrasts.
This work has been published in [22] , [23] .
A Retinotopic Neural Fields Model of Perceptual Switching in 2D Motion Integration
Participants : Pierre Kornprobst, Guillaume S. Masson [Institut de Neurosciences de la Timone, UMR 6193, CNRS, Marseille, France] , Kartheek Medathati, James Rankin.
In perceptual multistability a fixed but ambiguous stimulus can invoke multiple interpretations although only one can be held at a time. Visual motion stimuli are inherently ambiguous, for instance due to the aperture problem, which makes motion perception a complex inference task. The underlying cortical dynamics that select one percept out of multiple competing possibilities are not fully understood. Recent studies by [22] and [68] have tried to address this problem using the neural fields formalism. In [22] , a switching behaviour for a classical psychophysics stimulus, the multistable barberpole, was successfully captured in a feature-only, one-layer model of MT with adaptation and noise. However, without a representation of space, only some very specific stimulus could be considered. The work reported in [68] provides a much more general framework for motion integration in a two layer-model, however, it fails to capture the switching behaviour as the mechanisms of adaptation and noise were not considered. Building on the strengths of both studies, we propose a model that takes into account the spatial domain in a two-layer configuration whilst incorporating both adaptation and noise. Interactions between two layers processing local motion (V1 and MT) occurred through recurrent and lateral connections. The input stimuli are represented using direction of motion signals extracted using Reichardt detectors at corresponding 2D spatial locations. We use stimuli such as drifting bars and barberpoles to constrain the model to a suitable operating regime. In terms of computations, since the model is demanding, we implemented it using GPUs, extending the methods of [13] . Based on this implementation, we study dynamics of the model focusing on coherency in plaid motion (plaids and crossed barber pole).
This work has been presented in [28]