Abstract
Perceptual stimuli can be defined as parametric variations along multiple axes. Alternate parameterizations can partition the stimuli differently; for example a set of rectangles can be described as varying in height and width or in aspect ratio and size. It may be of interest to determine if a given pair of stimulus axes recruits neural populations that represent the two dimensions independently or conjointly. With fMRI, conjoint and independent representation may be distinguished through the examination of neural dissimilarity data that measure the inter-stimulus distance in neural representation between all pairings of a set of stimuli. Specifically, the identification of a Euclidean neural distance metric (Minkowski=2) versus City-block (Minkowski=1) for pairings across both of the perceptual dimensions indicates conjoint and independent coding, respectively. Using a continuous carry-over design (GK Aguirre, 2007), the necessary neural dissimilarity measures may be obtained on both focal (within voxel adaptation) and distributed (across voxel pattern analysis) scales. We will present simulations that suggest that adaptation and pattern metrics measured with fMRI reflect underlying conjoint and independent neural coding. We describe the construction of a “sub-additivity” covariate that may be used to estimate the Minkowski metric of neural representation, and the results of simulations that demonstrate the robustness of the approach to linear and non-linear distortions of the neural and hemodynamic responses. Optimizations of the stimulus space and the order of stimulus presentation within a continuous carry-over design are considered to maximize efficiency for estimation of the Minkowski neural metric. The results are further extended to consider non-metric cases (Minkowski [[lt]]1, “feature contrast”; and Minkowski = infinity, “single feature”). The results of application of these techniques to the study of two-dimensional object shape and aspects of facial appearance will be presented.