Abstract
The biased-competition theory accounts for the effects of attention at the neuronal level. Two hallmarks of this model are 1) the neuronal response to simultaneously presented stimuli is a weighted average of the response to isolated stimuli, and 2) attention biases the corresponding weights in favor of the attended stimulus. However, perception is not a property of single neurons, but probably relies instead on the activity of larger populations of neurons, which could be reflected in fMRI patterns of activity. Because several non-linearities can influence the pooling of single-neuron responses into BOLD signals, the fMRI effects of attention need not exactly mirror those observed at the neuronal level. Here we ask 1) how simultaneous stimuli are combined in multi-voxel patterns of representation and 2) how this effect depends on stimulus category, the brain region under consideration, and the allocation of attention. We considered data from an fMRI study in which four object categories (faces, houses, shoes and cars) were presented in four conditions: in isolation, or in pairs such that each category was attended, unattended, or attention was divided equally between the two. Unlike traditional analyses that collapse the response across all voxels in a region of interest, the response in each condition was represented in a multi-dimensional space where each voxel defined a dimension. In this high-dimensional space, the BOLD response to two simultaneously presented categories was well described as a weighted average of the response to individual stimuli. The weights were biased towards the preferred category in category-selective regions (FFA and PPA for faces and houses, respectively). Independently of this category-specific effect, and consistent with the biased competition theory, attention shifted the weights in favor of the attended stimulus, and the magnitude of this shift (30%) was quantitatively consistent with previous reports in single neurons.
This work was supported by grants from the Fondation pour la Recherche Medicale (L.R), ANR (project ANR 06JCJC-0154), Fyssen Foundation and EURYI (R.V), and EY13455 (N.K).