A Computational Model of Eye Movements as Cyclic Top-Down and Bottom-Up Processes

Abstract

Studying eye movement not only has theoretical significance in understanding visual attention, but also practical relevance to a diverse range of human factors applications such as aviation, driving, and display design. This article presents the use of a computational model of eye movement based upon reinforcement learning to examine the cyclic influences of top-down and bottom-up processes. The first study showed that different policies obtained from different goals in viewing a picture produced different types of eye movement patterns. The second study showed that the feedback information from each saccadic eye movement could be used to update the model's eye movement transition matrix, which led to different patterns in the subsequent saccade in a visual search task. These two studies demonstrate the value of an integrated reinforcement learning model in explaining both top-down and bottom-up processes of eye movement within one computational model.