Abstract
A body of research suggests that learning to categorize objects along a new dimension changes the perceptual representation of such dimension, increasing its discriminability and its separability from other dimensions. However, little is known about exactly how the internal representations of individual objects change during such dimension learning. Here, we trained twenty participants to categorize faces that varied along two morphing dimensions. One of the morphing dimensions was relevant to the categorization task and the other was irrelevant. We used classification images to estimate the internal templates used by participants to identify four faces varying along the category-relevant and category-irrelevant dimensions, both before categorization training and after categorization training. The obtained classification images provide estimates of the exact stimulus information used by the participants to identify the faces at each stage. Thus, examination and comparison of the obtained classification images allowed us to determine exactly how the internal representation of these faces changed as a result of categorization training. We defined two ways in which the representation of the category-relevant dimension could have changed as a result of categorization training. First, the internal templates of two faces having opposite values in the category-relevant dimension could become negatively correlated, a result that has been found with some familiar face dimensions. Our results suggest that categorization training had an effect in this direction, but the effect was not significant. Second, the internal templates of two faces having the same value in the category-relevant dimension, but different values in the category-irrelevant dimension could become more similar, which would explain previously-observed increases in dimensional separability after categorization training. Our results show a robust effect of categorization training in this direction.
Meeting abstract presented at VSS 2017