Model for Visual Decision Making Under Time Pressure.

摘要

This proposal examined how humans combine multiple sources of information when deciding to select a region of interest in a visual display. We considered tasks in which the observer was under time pressure to search a noisy display for a variable number of targets, or when the observer had to build an accurate representation of a new shape. We measured task performance as well as eye-scanning strategies as observers performed these tasks and compared these measures to the predictions of the optimal model. The model predicts that the most efficient strategy is to direct fixations to regions of maximum uncertainty, i.e. to regions that would maximize the information gained. Consistent with our model predictions, human eye movements in both tasks preferentially go to locations about which the observer is most uncertain. We also examined whether eye movements became more efficient when costs/penalties were involved. Performance improved slightly (5%) when observers were trained with a penalty for inefficient eye movements. A more effective way of improving performance was to guide humans to make eye movements to locations predicted by the optimal model. These results have implications for using the model both for long-term training and for on-line assistance to the human operator.