ImageNet representational similarity is algorithmic, not mechanistic, argues neuroAI thread

ShahabBakht · x · 2026-09-15

A follow-up in the same neuroAI thread: ImageNet-trained models share representational similarities with parts of the visual hierarchy, but the brain almost certainly didn't acquire these representations by solving supervised object categorization (or even its self-supervised variants). The similarities are algorithmic or computational findings, not mechanistic ones.

Related event: NeuroAI Critique: Brains Don't Learn via ImageNet-Style Classification(2 posts)→

Original post →

More from Research

Research channel →