Distributed representations of behaviour-derived object dimensions in the human visual system

How does the brain organize our understanding of the objects we see? A common view is that our brain separate objects into clear categories, such as faces or tools, with dedicated brain regions for each of these categories. Here, we asked whether the brain instead represents objects along continuous dimensions that matter for behaviour. While previous work has mapped some of these dimensions to the brain (like real-world size), a complete model of how the brain represents the multidimensional nature of objects has been missing.

We took dimensions derived from large-scale human similarity judgments (how people spontaneously group and distinguish objects) and mapped them onto brain activity across the brain using fMRI.

The result: behaviourally relevant dimensions are represented in a broadly distributed way across the human visual system, with mixed selectivity in many areas and sparser, dimension-specific tuning in some category-selective regions. These dimensions predicted brain responses better than category labels did.

The findings suggest that object vision is organized around continuous, behaviourally relevant dimensions. This framework also helps reconcile ongoing debates about seemingly opposing views of how the brain represents objects in terms of distinct regions vs. broader tuning maps. More generally, this work puts forth an alternative approach for understanding the brain, focusing less on presupposed categories or visual features and instead by staring with human behavior as a guiding principle for brain organization.