Dr Nikolaus Kriegeskorte, University of Cambridge
Human IT: a categorical and continuous representational space that predicts individual perception and enhances task-relevant divisions
Human object-vision fMRI has described category-selective regions and distributed category information. The field started out by focusing on brain activation averaged across image exemplars within each category, across voxels within cortical regions, and across individuals. Recent studies have engaged the complexities of single-image representations, distributed information, and individually unique representational geometries. I will present results suggesting that the hIT representational space (1) emphasizes fundamental categorical divisions more strongly than can be accounted for by visual-feature computational models (unless the features are explicitly optimized with supervised training), (2) despite consistent categorical clusters across individual brains, represents particular object images in a continuous space that is unique to each individual and predicts individual idiosyncrasies in object similarity judgements, and (3) flexibly emphasizes task-relevant category divisions through subtle distortions of the representational geometry.