A feedback model of visual attention

A model is presented to investigate the role of feedback pathways in attentional mechanisms. The model illustrates how feedback pathways help in dynamically changing the tuning properties of lower level neurons, thereby improving their convergence. These dynamic changes to the tuning properties also help in the recognition of a previously learned sequence of images. Two aspects of attention modeled are a gradual increase in concentration for finer convergence of the image, and dynamic shifting of the focus of attention to recognize a sequence of associated objects. Computer simulations of the system are presented. A nongradient-descent stochastic optimization algorithm was used in the simulations. The system is capable of cycling through patterns whose associations have been learned in higher-level representations.<<ETX>>