It is found that a common result here -- that untrained or locally trained networks rival or beat backpropagation at early visual cortex -- depends strongly on the resolution at which the network is evaluated.
Abstract
Studies of biologically plausible learning rules (feedback alignment, predictive coding, STDP) train small networks on $32\times32$ 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 $-0.001$ (95% CI $[-0.011, 0.009]$) at the 32 px training resolution to $+0.030$ $[0.018, 0.042]$ 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 ($+0.035$ $[0.023, 0.046]$ vs. $+0.030$ at 224 px; ratio 1.14 $[1.03, 1.34]$): 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.
It is proposed that CNN training's classification bottleneck compresses brain-relevant information at depth, unlike transformers's self-attention and non-classification objectives, which reflect signal strength and persistence rather than distinct brain regions.
S. Baghel, Kshitij Dwivedi, Dinesh Singh et al.· arXiv.org· 0 citations
How neural activity across the ventral visual hierarchy supports face recognition is an open question. A long-standing debate asks whether face processing, particularly in fusiform cortex, relies on face-specific computations or representations shared with broader visual recognition. Here we combine source-resolved mag...
Hamza Abdelhedi, Shahab Bakhtiari, Karim Jerbi· bioRxiv· 0 citations
These results show that nested neural contrasts do not identify represented content by themselves: predictor construction is part of the experimental design, and matched controls are required for representational claims.
Lucas Nadolskis, Galen Pogoncheff, Michael Beyeler· 0 citations
This work introduces axis-aligned feature accentuation, which converts each model’s fitted encoding axis into graded stimulus perturbations that are predicted to parametrically control neural firing within and beyond the natural-image range.
Jacob S. Prince, Binxu Wang, Thomas Fel et al.· bioRxiv· 0 citations
Visual perceptual learning is often highly specific to the trained retinal location, a hallmark observation that is traditionally interpreted as evidence that training-induced plasticity occurs in retinotopic early visual cortex. However, under properly designed training conditions, such as double training, adaptation...
Lin Zhong, Gong-Liang Zhang, Jun-Yun Zhang et al.· NeuroImage· 0 citations
Functional digital twins found that when and whether a surround facilitates or suppresses the optimal center feature depends on how congruent the surround is given that feature under natural scene statistics, consistent with hierarchical Bayesian inference.
H. Mansvelder, A. Kreiter, Ziad M. Hafed et al.· 0 citations
Related blog posts
Microsoft Research Blog· microsoft.comNov 16, 2026
AI coding agents are increasingly help developers fix bugs and build features largely on their own, but every interaction with their models consumes tokens. In each session, an agent sends instructions, tool descriptions, and session history to its model. Most efforts to reduce this cost focus on individual tasks. Yet developers return to the same […] The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduNov 13, 2026