Abstract
The neural response to visual stimuli is typically attenuated when similar or identical items have been recently encountered—an effect known as neural adaptation. It has been suggested that adaptation may serve to decorrelate neural responses across the set of representational channels. Empirical evidence for this idea, however, is sparse. Here we used multivoxel pattern analysis to test whether adaptation induces decorrelation for object representations. We constructed a parametric 3D shape-space whose axes corresponded to different frequencies and orientations of sinusoidal modulations of a sphere. Two distinct reference shapes (A and B) were selected from this space, along with a “cloud” of fifty stimuli around each reference. BOLD fMRI data were collected over 8 runs. Each run started with a pre-adaptation phase (60s), where subjects were presented with an RSVP stream of stimuli drawn randomly from one of the clouds (cloud A in 4 runs; cloud B in 4 runs; counterbalanced). Afterwards, each of the 16 trials started with a top-up adaptation phase (5s) followed by a test stimulus (1s) at a jittered (4-7s) ISI. The test stimuli throughout all runs were four items, two drawn from cloud A, and two drawn from cloud B. To quantify decorrelation, we extracted the patterns for these four items in each scan run and performed a similarity analysis. Results indicate that the patterns evoked by two similar shapes (e.g. A1 and A2) were less correlated when these shapes were preceded by stimuli from the corresponding cloud than when they were preceded by stimuli from the other cloud, in both anterior and posterior portions of the Lateral Occipital Complex (defined by an independent functional localizer). These results demonstrate that adaptation decorrelates representations in object-selective regions.
Meeting abstract presented at VSS 2015