q-bio.NCNov 11, 2026

Analysis choices decide whether untrained networks beat trained ones at V1: a multiverse analysis of model-brain RSA

Authors: Nils Leutenegger

Organizations: Independent researcher, Zug, Switzerland

Abstract

Untrained convolutional networks have been reported to rival trained networks as models of visual cortex, and we had reported an untrained-network advantage over backpropagation at V1 in THINGS fMRI, together with a dependence on evaluation resolution. Under a pre-specified ridge readout, the untrained network scored below the trained one at every pooling grid (median r difference -0.043 at g = 4, 0 of 5 seeds positive), and also in MSE skill and per-fold r. In exploratory analyses, across 36 combinations of unit normalisation, distance measure and voxel normalisation (28 distinct analyses), the sign of the V1 contrast depended on the analysis (intervals above 0 in 5 and below 0 in 9 distinct analyses), and in our networks the resolution dependence of the original convention appeared in only one cell. Raw correlation-distance representational similarity analysis (RSA) covaried with the alignment of model and voxel patterns with the unit-mean profile (Spearman 0.97 across our networks, 0.79 across six ImageNet architectures); the advantage was absent from the component orthogonal to the mean profile, and the parallel component alone did not resolve it. On the unweighted angular component of centred patterns, trained networks were ahead at V1 (our network and 5 of 6 ImageNet models) and at LOC (all seven); the norm-weighted angular term was not resolved. A luminance-and-contrast bank scored above every network under ridge (point estimates), and adding network features under a shared penalty did not improve prediction. Intervals are stimulus-bootstrap intervals for three subjects, and 6 pre-stated predictions were not met. We recommend reporting sensitivity to centring over the evaluated stimuli on both sides, reporting components separately, and including a small multiverse and simple baselines.

Figures & tables

Appendix figures & tables13 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Aug 11, 2026

Evaluation Resolution Confounds Learning-Rule Comparisons in Model-Brain RSA of Early Visual Cortex

Revision note: see Comments. Studies of biologically plausible learning rules (feedback alignment, predictive coding, STDP) train small networks on 32×3232\times32 images and compare them with brain responses to naturalistic stimuli presented at much higher resolution. We find that a common result in this setting, that untrained or locally trained networks rival or beat backpropagation at early visual cortex, depends on the resolution at which networks are evaluated. In representational similarity analysis (RSA) against human V1, the gap between an untrained and a backprop-trained network grows monotonically from −0.001-0.001 (95% CI [−0.011,0.009][-0.011, 0.009]) at the 32 px training resolution to +0.030+0.030 [0.018,0.042][0.018, 0.042] at 224 px (six resolutions, 5 seeds; per-subject RSA on cross-run stimulus pairs). The growth reproduces for three further learning rules, directionally in single-seed macaque electrophysiology, along training, and for an ImageNet ResNet-50 and a Swin-Tiny transformer, which also align best at low resolution although trained at 224 px. Gabor and pixel structure, the normalization state of the untrained baseline, and a global brightness statistic do not account for it. Against a criterion fixed before the analysis, limiting image content to 32 px while the input still grows does not shrink the V1 gap (+0.035+0.035 [0.023,0.046][0.023, 0.046] vs. +0.030+0.030 at 224 px; ratio 1.14 [1.03,1.34][1.03, 1.34]): it arises once the input exceeds the training resolution, through a decline of backprop. A single luminance value per image reaches rho= 0.054 against V1, matching the untrained network (0.053), which bounds what RSA on globally pooled features can resolve at V1 in this dataset. The one learning effect that holds across resolution is backprop above untrained at LOC. Comparisons of learning rules or architectures at early visual cortex need to control, and report, the evaluation resolution.
Apr 18, 2026

Untrained CNNs Exceed Backpropagation in V1 Alignment at High Evaluation Resolution: A Systematic RSA Comparison of Four Learning Rules Against Human fMRI

Revision note: the V1 result reported here is not robust to the analysis choices examined in arXiv:2611.15533; see page 1. We compare four learning rules (backpropagation (BP), feedback alignment (FA), predictive coding (PC), and spike-timing-dependent plasticity (STDP)) in identical convolutional networks against human fMRI from THINGS-fMRI (720 stimuli, 3 subjects) using representational similarity analysis (RSA), together with an untrained random-weights baseline that isolates the contribution of architecture. Models are trained on 32 px CIFAR-10 and evaluated at 224 px; results are averaged across 5 seeds, with uncertainty from resampling stimuli. At V1, the untrained baseline exceeds BP rho = 0.053vs.vs.0.022;rho=+0.031; rho = +0.031, 95% CI [0.019,0.043][0.019, 0.043]), and the same holds at V2. This advantage depends on the evaluation resolution: in the 5-seed model set of the resolution study (gap +0.030+0.030 at 224 px), it vanishes when the models are evaluated at their 32 px training resolution rho = -0.001, 95% CI [−0.011,0.009][-0.011, 0.009]; arXiv:2608.12408). At LOC, BP, PC and STDP exceed the untrained baseline, FA is borderline, and the trained rules do not differ from each other. At IT, no two conditions can be distinguished at the low reliability of the data (leave-one-subject-out lower bound 0.021). FA has the lowest alignment at V1. At V1, an untrained network thus matches BP at the training resolution and exceeds it at 224 px; learning-rule differences are resolvable only at an intermediate area.
May 28, 2026

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

CORRECTION (August 2026): the central finding of this paper is not supported. An evaluation-mode defect left the batch-normalisation layers of the predictive-coding and STDP conditions in training mode during feature extraction, producing their apparent preservation of V1 alignment. With the defect repaired, predictive coding degrades V1 alignment more than backpropagation does, not less. The finding that training degrades V1 alignment for every rule tested does survive. [Revision note: not robust to the analysis choices examined in arXiv:2611.15533.] See the correction note on page 1; the original abstract below and the body are unchanged from v1. Corrected analysis: arXiv:2608.12408. Revision note: see Comments. Random, untrained neural networks consistently match or exceed trained networks in representational similarity to early visual cortex. This puzzling finding challenges the assumption that learning improves brain alignment. We investigate it by tracking representational similarity analysis (RSA) alignment to human fMRI data across training for four learning rules: backpropagation (BP), feedback alignment (FA), predictive coding (PC), and spike-timing-dependent plasticity (STDP). Using 720 object images from the THINGS database and fMRI data from three subjects across six visual ROIs, we measure Spearman correlations between model and brain representational dissimilarity matrices at eight training checkpoints (epochs 0-40). We find that (1) a single epoch of training reduces V1 alignment by 25-90%, depending on the learning rule; (2) backpropagation reduces V1 alignment most severely (delta r = -0.080), while predictive coding and STDP preserve substantially more (delta r ~ -0.04); and (3) a weaker, opposite tendency appears in object-selective cortex (LOC), where BP shows the largest increase in alignment during training, although the absolute change is small.