Across-Trial Statistical Learning in Visual Search

This dissertation examined the possibility for observers to learn spatial associations that occurred across consecutive search displays and to use these regularities to improve their search efficiency. In Chapter 2, we conducted four experiments in which participants were asked to search for a shape singleton target. As a result, we consistently observed that the search performance on predictable trials was better than that on unpredictable trials. The results suggest that participants were able to implicitly pick up across-trial regularities regarding target locations both in the absence (Exp. 1 and 2) and presence (Exp. 3 and 4) of a color singleton distractor. In Chapter 3, we investigated whether there is across-trial learning of target locations when the search is serial, employing a T-among-Ls task. We showed that across-trial learning on target locations occurred only when these targets were salient enough to be immediately selected (Exp. 2), but not when these targets were non-salient (Exp. 1). In Chapter 4, we examined whether across-trial regularities related to salient (pop-out) but task-irrelevant distractor locations could be learned and bias attentional selection via suppression in visual search. Across five experiments, we demonstrated that participants were unable to pick up and use across-trial regularities regarding the salient distractor location in visual search. In Chapter 5, we explored the role of responding in across-trial learning regarding target locations in pop-out feature search. Combining the paradigm of Exp. 1 in Chapter 2 with a Go/NoGo task, we showed that across-trial target location learning occurred only when responding was required to both targets of a pair, not when the response to the target within the pair needed to be withheld (Exp. 1). Additionally, the absence of learning in Exp. 1 cannot be attributed to the carry-over inhibition resulting from not having to response on the preceding trial (Exp. 2). In Chapter 6, using EEG, we showed the existence of dynamic weight changes within the assumed spatial priority map, in a paradigm in which participants were exposed to a sequence of dots. After an initial exposure phase, partial sequence trials, wherein only one single dot was presented, were intermixed with full sequence trials. With multivariate pattern analyses trained on independent pattern estimators, we were able to decode not only the present dot in partial sequences but also the subsequently anticipated but omitted positions at their expected moment in time. The findings suggest that the brain is able to dynamically prioritize and internally generate representations for anticipated locations in a future-oriented manner, in the absence of sensory inputs. In sum, this dissertation demonstrated that observers are able to learn across-trial regularities and use them to guide spatial attention in visual search. Several boundary conditions for across-trial learning have been unveiled. Through across-trial statistical learning, the weights within the assumed spatial priority map can be dynamically adjusted from trial to trial. The findings of this dissertation contribute to understanding whether, when, and how across-trial regularities bias attentional selection in visual search.