Abstract
Creating associations between different objects is a critical way in which we interact with our environment. Forming new associations relies on the medial temporal lobe, while their longer-term storage is mediated by cortex. However, the large-scale neural changes accompanying these newly formed associations are not fully understood. In this study we investigated whether the fMRI multi-voxel representations of associated object categories become more similar to each other as a result of creating arbitrary new associations between them. Nine human subjects were scanned in a 3T scanner before and after they learned arbitrary associations between pairs of different object categories (faces, houses, cars, chairs). The learning procedure consisted of 20-minute sessions over 15 days during which subjects were required to learn arbitrary associations between 10 different exemplars of these categories (e.g., each face was associated with a car/each house with a chair). During the fMRI sessions before and after learning, subjects were presented with these categories in a blocked design and were required to perform a one-back task on the images. To evaluate how object representations change as a result of learning we used the searchlight method of analysis in which a classifier was trained to discriminate between categories A and B, and tested on discriminating C versus D, where A-C and B-D indicate the category pairs (arbitrarily) associated during learning. Our results show that cross-classification performance was at chance levels before learning (because the association was arbitrary, and the category assignment counterbalanced across subjects), but that after learning different regions of the occipital and temporal lobes displayed increased performance. Overall subjects and all voxels the average increase (2%) was highly significant (p<10-6). These results thus suggest that the representations of associated categories in these regions are not static, but become more similar to each other as a result of learning.
Meeting abstract presented at VSS 2014