Computational models of category-selective brain regions enable high-throughput tests of selectivity

Cortical regions selective to faces, places, and bodies have provided evidence for domain-specific theories of human cognition. This research develops artificial neural network-based encoding models that accurately predict responses to novel images in these brain areas, outperforming descriptive models and experts. These models were used to rigorously test category selectivity by screening for and synthesizing images predicted to produce high responses. The high-response images were all unambiguous members of the hypothesized preferred category, strengthening evidence for domain specificity in the brain and enabling future research with unprecedented computational precision.

N. Apurva Ratan Murty, Pouya Bashivan, Alex Abate, James J. DiCarlo, Nancy Kanwisher