Category learning in the visual processing stream

Abstract

Previous work has shown that category learning of dot patterns does not transfer to other retinal locations (Sinha, Heindel & Welch, 1998; Silverman & Welch, 2003 VSS meeting). In addition, learning 2D line drawings of 3D shapes also does not transfer much to other retinal locations (Welch, Boachie [& Silverman], 2003 VSS meeting). This lack of transfer suggests that the shape learning is happening early enough in the visual processing stream that the cells' receptive fields are relatively small. It might be possible to tap into a higher processing level if the training characteristics were modified to enhance learning at multiple locations. We trained subjects in the same category-learning paradigm as used previously but included two learning locations instead of one. We asked whether this would increase subjects' abilities to recognize shapes in untrained locations. We also used the 2D line drawings of 3D shapes because they showed a small amount of transfer to close by locations Surprisingly, subjects showed no more location transfer than what was seen with a single training location. Even a testing location between the two training locations showed no more transfer than any other location. Increasing variability in the training locations had no impact on learning transfer to untrained locations. The result is surprising because it suggests that learning to categorize line drawings of 3D shapes happens earlier in the visual processing stream than cortical areas that generally have been associated with object recognition such as AIT where cells' receptive fields are very large. Brown University