Untrained CNNs Match Backpropagation at V1: A Systematic RSA Comparison of Four Learning Rules Against Human fMRI
This work clarifies the region-specific roles of architecture and learning rules in aligning neural network representations with the human visual cortex, showing that early visual areas are dominated by architecture while higher areas require supervised objectives.
The study compares four learning rules (BP, FA, PC, STDP) against human fMRI data and finds that early visual alignment (V1/V2) is primarily driven by architecture, with an untrained CNN achieving rho=0.071, statistically indistinguishable from BP (rho=0.072, p=0.43). Learning rules only differentiate at higher visual areas, where BP and PC with local Hebbian updates achieve similar IT alignment.
A central question in computational neuroscience is whether the learning rule used to train a neural network determines how well its internal representations align with those of the human visual cortex. We present a systematic comparison of four learning rules -- backpropagation (BP), feedback alignment (FA), predictive coding (PC), and spike-timing-dependent plasticity (STDP) -- applied to identical convolutional architectures and evaluated against human fMRI data from the THINGS-fMRI dataset (720 stimuli, 3 subjects) using Representational Similarity Analysis (RSA). Crucially, we include an untrained random-weights baseline that reveals the dominant role of architecture. We find that early visual alignment (V1/V2) is primarily architecture-driven: an untrained CNN achieves rho = 0.071, statistically indistinguishable from BP (rho = 0.072, p = 0.43). Learning rules only differentiate at higher visual areas: BP dominates at LOC/IT, and PC with local Hebbian updates achieves IT alignment statistically indistinguishable from BP (p = 0.18). FA consistently impairs representations below the random baseline at V1. Partial RSA confirms all effects survive pixel-similarity control. These results demonstrate that the relationship between learning rules and cortical alignment is region-specific: architecture determines early alignment, while supervised objectives drive late alignment.