The visual system, like the brain more broadly, relies heavily on categorical representations. It is easier to spot a visual difference that crosses a category boundary, e.g., between blue and green, or between vertical and oblique. Here we show that topological relations between objects are similarly categorical. When asked to detect changes between object arrangements, participants were better at detecting those changes that crossed hypothesized category boundaries, such as ’overlapping’, or ’touching’, compared to equally-sized changes that did not. These effects were magnified at increased memory load, presumably because this categorization forms a more efficient code. This finding, consistent with previous computational modeling work, suggests that categorical relations are critical for remembering and comparing complex images.

This paper presents a theoretical framework for modeling human visual attention. The framework's core claim is that three mechanisms drive attention: selection, which picks out an item for further processing; engagement, which tags a selected item as relevant or irrelevant to the current task; and enhancement, which increases sensitivity to task-relevant items and decreases sensitivity to task-irrelevant items. Building on these mechanisms, the framework is able to explain human performance on attentionally demanding tasks like visual search and multiple object tracking, and it supports a broad range of predictions about the interactions between such tasks.

We present a computational model of attentional capture inhumans. The model distinguishes between automatic mecha-nisms that directly determine the focus of visual attention, anddeliberate mental actions an individual can perform to influ-ence these mechanisms. The automatic mechanisms select anobject as the focus of attention and enhance its location andfeatures, so that nearby or similar objects are likely to be se-lected in the future. The deliberate actions include engagingwith a selected object to further enhance its features, and re-trieving a previously selected object from memory. By per-forming these actions, the model is able to exert limited top-down control over capture, increasing the probability that task-relevant objects will be attended and irrelevant objects will beignored. To evaluate the model, we conduct a simulation of arecent visual search study, demonstrating that the model canaccount for three established factors that are known to influ-ence capture.

We present a computational model exploring goal-directeddeployment of attention during object tracking. Once selected,objects are tracked in parallel, but serial attention can bedirected to an object that is visually crowded and in danger ofbeing lost. An attended object’s future position can beextrapolated from its past motion trajectory, allowing theobject to be tracked even when it is briefly occluded. Usingthe model, we demonstrate that the difficulty of trackingthrough occlusions increases with the number of objectsbecause they compete for serial attention.