Developmental emergence of brain–model alignment

Determine how the representational alignments indexed by the encoding, spatial, and temporal scores between self-supervised vision transformers and human brain activity arise across development, rather than solely in the adult brain.

Background

Analyses in the paper are conducted on adult fMRI and MEG datasets, revealing that alignment emerges during model training with distinct timescales across cortical regions and MEG latencies.

The authors note that understanding when and how these alignments appear over human development requires data beyond adult cohorts, motivating a developmental investigation of the same alignment metrics.