engleski

Multiple-object tracking in a neural network for visual working memory

We proposed a recurrent neural network for visual working memory which is able to track objects as they move in the visual field. The model is based on queuing networks previously used in neural models of serial order in working memory. The proposed model consists of self-excitation and two types of inhibition: recurrent (or lateral) and dendritic inhibition. The model explicates how pointers to different objects are labeled with different activity levels and how this activation is updated as objects move in space. Computer simulations showed that the capacity to track objects is limited due to the lateral inhibition and to the restricted dynamic range of model neurons. However, capacity could be dynamically modulated by the object’s speed. Self-excitation enables tracking of objects which disappear and reappear behind visible or invisible occluded surfaces. When moving objects form a perceptual group, the capacity to track objects increased due to weak lateral inhibition among the members of the group. On the other hand, the model showed reduced capacity to track substances which extends in space as they move. Therefore, the model provides unified account of variability in capacity for multi-object tracking.

pažnja; praćenje objekata; neuronski model

nije evidentirano