[Article] What can 5.17 billion regression fits tell us about artificial models of the human visual system?
Summary: They performed a large-scale benchmarking analysis of 85 modern deep neural network models (e.g. CLIP, BarlowTwins,Mask-RCNN). They found that architectural differences have very little consequence in emergent fits to brain data. Next, differences in task have clear effects–with categorization and self-supervised models showing relatively stronger brain predictivity.