Expanding the Computational Capabilities of Guided Search with a Multidimensional Scaling "Plug-In"

Guided Search (GS) has been an effective computational model for studying visual search involving rudimentary stimuli, such as lines of various colors and orientations. The visual qualities of such stimuli can often be operationalized based on wavelength of light or degree of rotation, but this kind of quantification is not as straightforward for more complex visual stimuli, such as images of real-world objects. In the current investigation, we propose a method to use a multidimensional scaling (MDS) analysis of search stimuli as input into a modified GS model. To test the effectiveness of the proposed GS "plug-in," we conducted two experiments in which participants searched for novel objects among distractors that varied in how similar they were to the target and to each other (defined by distance in MDS space). Search was more efficient when distractors were dissimilar from the target compared to when they were similar to the target. Modeled data from the modified GS model effectively captured trends in target-present responses. This analysis supports the use of MDS to model similarity and suggests that it may be a viable tool for extending computational uses of GS, as well as for further investigating target-absent responses within the GS framework.