Patrick Mineault Examines Four Differences Between the Visual Cortex and Artificial Neural Networks, Challenging the Platonic Representation Hypothesis

Written by

in

Patrick Mineault identifies four differences between the visual cortex and artificial neural networks: prediction, geometry, linearization, and robustness. This suggests that the Platonic representation hypothesis does not fully hold. Existing visual ANNs explain only about half of the variance in IT neurons in the high-level visual cortex, while models with top-1 accuracy above approximately 70% show lower neural alignment. The author recommends exploring digital twin models trained directly on neural data to determine whether these differences can be bridged, although existing neural datasets remain limited in size.

Source:The NeuroAI archive · Read the original

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *