Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Representational tuning models: Uniting representational similarity analysis and neural encoding models.

Stan Bergey¹, Jorge Almeida, Luis Martín-Roldán Cervantes, Ben Harvey; ¹Tilburg University

Presenter: Stan Bergey

In fMRI analysis, neural encoding models reveal how individual voxels respond as a function of continuous stimulus parameters, and how response function parameters change within and between brain areas. Conversely, representational similarity analysis reveals structure in the responses of a region of interest (ROI) to any stimuli. Here we develop representational tuning models to unite these approaches. These response models first rescale the representational dissimilarity matrix (for an ROI) to a 2-dimensional representational space then find (for each voxel) the Gaussian function within this 2D space that best predicts the voxel’s response to all stimuli. By deriving continuous response function parameters from the ROI’s responses, this approach requires no a priori hypothesis of stimulus parameters underlying the response function. It thereby allows application of neural encoding models to arbitrary stimulus sets. We test this approach for responses from the Natural Scenes Dataset within 12 visual field maps. We show that representational tuning models significantly predict voxels’ responses to natural images in higher-level (but not early) visual field maps, especially when sampling from other visual field maps, and we demonstrate that the principal components of representational spaces reflect the spatial structure of responses across the cortical surface within an ROI.

Topic Area: Methods & Computational Tools

Extended Abstract: Full Text PDF