A Neural Network Model for Gating Task-Relevant Information by Rhythmic Oscillations

Abstract

Visual system processes simple object features in early visual areas, and visual features become more complex as visual information is sending to downstream areas. In addition to the feedforward pathway, visual system has abundant feedback connections, whose number is even larger than feedforward ones. This suggests that top-down signal from higher visual areas may strongly affect sensory representation of early visual areas. Also, visual processing along the feedforward and feedback pathways is coordinated by brain rhythms. However, little is known about how the bidirectional visual processing is related with brain rhythms. To address this issue, we focus on an experimental study using two tasks in a visual perception. We develop a model of visual system which consists of a V1 and a V2 network. Using the model, we show that tuning modulations of V1 neurons are caused by a top-down influence mediated by the change in long-range connections of V1 neurons. We also show that top-down signal reflecting a slower oscillation in V2 neurons, coupled with a fast oscillation of V1 neurons, enables the efficient gating of task-relevant information encoded by V1 neurons.