Self-supervised learning yields representational signatures of category-selective cortex.

Self-supervised neural networks develop category-selective features similar to human brain regions, with 80% of key signatures matching those in the FFA and PPA.

• Why it matters: Understanding how the brain's category-specific areas acquire their distinct tuning can reveal fundamental principles of visual processing and inform artificial intelligence development.
• What they did: Researchers applied a functional localizer approach to both humans and self-supervised models, identifying face- and scene-selective units and comparing their responses across various visual features.
• The result: Findings show that domain-general learning objectives are enough to produce human-like category selectivity, implying that specialized brain regions may emerge from a unified, self-supervised computational goal.