A Computational Model to Study the Dynamics of Updating of Remembered Visual Targets During Rapid and Slow Eye Movements

Abstract

After an intervening eye movement, or saccade, humans and animals are able to localize previously perceived visual targets (spatial updating). Although efforts have been made to discover the mechanism underlying spatial updating, there are still many unanswered questions about the neuronal mechanism of this phenomenon. State space model is an effective method for modeling dynamical systems and it can represent the internal behaviour of these systems. Therefore, we developed a state space model for updating target-related spatial information in gaze-centered coordinates. We considered three types of input in our proposed model: 1) an efference copy signal, inspired by motor burst in SC, 2) an eye position signal, found in LIP, VIP, MT and MST areas and 3) visual topographic maps of visual stimuli, located in SC. To model the internal neuronal behaviour of the system, we developed a radial basis function neural network (RBFNN) which can be trained with an Extended Kalman filter method. This RBFNN represents the state space and we can obtain a topographic map of the remembered target in its hidden layer. From our proposed model, the output obtained is the decoded location of the remembered target. To explore the internal mechanism underlying the updating process, we trained this model on a double-step saccade-saccade or pursuit-saccade task. After training, the receptive fields of state-space units replicated both predictive remapping during saccades (Duhamel et al. Science 1992) and continuous eye-centered updating during smooth pursuit (Dash et al. Current Biology, in press). In addition, during trans-saccadic remapping, receptive fields also expanded (to our knowledge, this predicted expansion has not yet been reported in the published literature). In the future, we plan to incorporate this framework within a more comprehensive model of trans-saccadic integration of both spatial and feature information, and use this framework to construct a physiologically plausible model. Meeting abstract presented at VSS 2015. Language: en