aiqbench analysis
This study investigates whether untrained random weights in convolutional neural networks can match the representational similarity of trained models to human visual cortex data at V1, finding that architecture alone may drive alignment more than specific learning rules. A significant correction regarding evaluation-mode defects invalidates previous claims for predictive coding and STDP conditions while confirming robustness for backpropagation, feedback alignment, and random baselines across different cortical areas. The findings highlight the critical role of network structure over training dynamics in mimicking early visual processing stages without explicit supervision.
Untrained CNNs Match Backpropagation at V1: A Systematic RSA Comparison of Four Learning Rules Against Human fMRI
arXiv:2604.16875v3 Announce Type: replace Abstract: CORRECTION (August 2026): an evaluation-mode defect affected the predictive-coding and STDP conditions of this study; those results should not be used pending re-computation. At V1 and 224px, predictive coding falls from rho = 0.056 to 0.016 and STDP from 0.064 to 0.037, so the claims that STDP leads among trained rules and that PC and STDP lead at V1/V2 are not supported. The random, backpropagation and feedback-alignment conditions, which carry the untrained-versus-trained claim, are unchanged to within 0.0013. The result is however strongly dependent on the evaluation resolution, held fixed at 224px here; see arXiv:2608.12408. See the correction note on page 1; the original abstract below and the body are unchanged from v1. 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). All models process stimuli at 224 x 224 resolution; results are averaged across 5 random seeds. Crucially, we include an untrained random-weights baseline that reveals the dominant role of architecture. At V1/V2, the untrained baseline exceeds backpropagation (rho = 0.076 vs. rho = 0.034; Delta-rho = +0.044, p < 0.001). At LOC, only BP reliably exceeds the random baseline (rho = 0.012 vs. -0.005, p < 0.001). At IT, all five conditions converge (rho = 0.008-0.014) with no significant pairwise differences among trained rules. Partial RSA confirms all effects survive pixel-similarity control.
Abstract:CORRECTION (August 2026): an evaluation-mode defect affected the predictive-coding and STDP conditions of this study; those results should not be used pending re-computation. At V1 and 224px, predictive coding falls from rho = 0.056 to 0.016 and STDP from 0.064 to 0.037, so the claims that STDP leads among trained rules and that PC and STDP lead at V1/V2 are not supported. The random, backpropagation and feedback-alignment conditions, which carry the untrained-versus-trained claim, are unchanged to within 0.0013. The result is however strongly dependent on the evaluation resolution, held fixed at 224px here; see arXiv:2608.12408. See the correction note on page 1; the original abstract below and the body are unchanged from v1. 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). All models process stimuli at 224 x 224 resolution; results are averaged across 5 random seeds. Crucially, we include an untrained random-weights baseline that reveals the dominant role of architecture. At V1/V2, the untrained baseline exceeds backpropagation (rho = 0.076 vs. rho = 0.034; Delta-rho = +0.044, p < 0.001). At LOC, only BP reliably exceeds the random baseline (rho = 0.012 vs. -0.005, p < 0.001). At IT, all five conditions converge (rho = 0.008-0.014) with no significant pairwise differences among trained rules. Partial RSA confirms all effects survive pixel-similarity control.
11 pages, 9 figures. v2: Correction note added (p.1). An evaluation-mode defect affected the predictive-coding and STDP conditions; those results should not be used pending re-computation. Random, backprop and feedback alignment are unaffected. See also arXiv:2608.12408
Machine Learning (cs.LG); Neurons and Cognition (q-bio.NC)
arXiv:2604.16875 [cs.LG]
(or arXiv:2604.16875v3 [cs.LG] for this version)
Full text under the publisher's open licence · 332 words
Related coverage
Evaluation Resolution Confounds Learning-Rule Comparisons in Model-Brain RSA of Early Visual Cortex
arXiv · Aug 25, 2026
Sakana AI’s Error Diffusion Trains Dale-Compliant Dual-Stream Networks, Reaching 96.7% MNIST and 61.7% CIFAR-10 Without Backpropagation
MarkTechPost · Jul 18, 2026
The AI Inference Revolution Is Here
IEEE Spectrum · Sep 15, 2026
Model Weight Profiles: Where Do the Parameters Go?
AMD · Sep 25, 2026