MS Thesis: A Visual search task to evaluate top-down and bottom-up control of the pre-attentive stage and the ACT-R/PM vision module

Hakan Unlu, June 2006, Bogazici University, Cognitive Science MS Program



Visual Attention is deployed in two stages: The pre-attentive stage determines which areas of the visual field are relevant for the task and therefore need to be attended. The attentive stage processes the visual information available at the attended portion of the visual field. Two rival views suggest that the pre-attentive stage is controlled by physical properties of the visual field (bottom-up) or the goals and intentions of the observer (top-down). In support of the bottom-up approach, Theeuwes conducted an experiment to show an irrelevant singleton cannot be masked in a top-down fashion. However Bacon and Egeth (1994) suggested that the nature of the task dictates which method will be used. In this study, three experiments were conducted to test Theeuwes’ Irrelevant Singleton hypothesis and Bacon and Egeth’s Feature Search hypothesis. The results were not compatible with either claim. The experiment results are further analyzed. Data indicate that, (i) search times depend on the color, location, set size and the form (ii) the time spent per item is larger when there is no target in the display; (iii) in the presence of a target, the average search time per item is inversely proportional to the set size. Several possible explanations are discussed.

ACT-R/PM is a cognitive architecture that allows a cognitive task to be modeled in computer environment. One of our experiment setups was modeled in ACT-R/PM to verify that ACT-R/PM can model our task. The results show that, when default parameters are used, ACT-R/PM is slower than human participants. Also, ACT-R models fail to show the inverse relation between average response time per item and the set size. These results were evaluated and a criticism of current ACT-R/PM constructs was provided.

Leave a Reply

%d bloggers like this: