Supervised Training Rapidly Degrades Early Visual Cortex Alignment Across Biologically Plausible Learning Rules

In three linesUntrained neural networks match early visual cortex better than trained networks. Study on 720 THINGS images and fMRI from 3 subjects shows one training epoch reduces V1 alignment by 25-90% depending on learning rule. Backpropagation degrades most (Δr = -0.080), while predictive coding and STDP preserve alignment better (Δr ~ -0.04).

Summary generated by Claude — human-verified