A Connectionist Model of Visual Attention

点击翻译按钮获取中文摘要

Selective visual attention serializes the processing of stimulus data to make efficient use of limited processing resources in the human visual system. This work will develop a connectionist network to emulate a variety of attentional phenomena that have been reported by Treisman, Wolford, Duncan and others. As demonstrated in preliminary work, a hierarchical, multi-scale network that uses feature arrays with strong lateral inhibitory connections is capable of producing responses that are suggestive of a number of key behaviors associated with visual attention. The work to be done involves integrating the network structures of the earlier work into a single cohesive architecture, elaborating this network model to support additional function, and tuning the network structure to provide a more quantitative match with behavioral and neural data. The goals of the work are to provide (1) a practical computational model for machine recognition systems, (2) a model of human visual attention that is more comprehensive than current theories, and (3) a network model that makes contact with data on the neural basis of attention.