Universal dimensions of visual representation
Open accessThe DOI matches its Crossref record; 45 of the 45 references it cites were found in a registry or an index.
Abstract
Do visual neural networks learn brain-aligned representations because they share architectural constraints and task objectives with biological vision or because they share universal features of natural image processing? We characterized the universality of hundreds of thousands of representational dimensions from networks with different architectures, tasks, and training data. We found that diverse networks learn to represent natural images using a shared set of latent dimensions, despite having highly distinct designs. Next, by comparing these networks with human brain representations measured with functional magnetic resonance imaging, we found that the most brain-aligned representations in neural networks are those that are universal and independent of a network’s specific characteristics. Each network can be reduced to fewer than 10 of its most universal dimensions with little impact on its representational similarity to the brain. These results suggest that the underlying similarities between artificial and biological vision are primarily governed by a core set of universal representations that are convergently learned by diverse systems.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Figures





Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“These empirical findings have led to the theoretical proposal of representational Universality: different information-processing systems (ANNs and brains) all converge upon the same shared representational structures because there exist ‘general-purpose dimensions that have little to do with the details of a network’s architecture or task objective’ [12]”
From The Umwelt Representation Hypothesis: rethinking Universality (2026) · cites this paper as [12]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“However, recent results suggest that contrasting different ANNs with neural data may start to yield diminishing returns: models with drastically different architectures and training regimes all match neural data with similar accuracy, especially when parametric flexibility is allowed in the model–brain mapping [6,12–14]”
From The Umwelt Representation Hypothesis: rethinking Universality (2026) · cites this paper as [12]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Furthermore, recent studies have found a striking representational alignment between a wide variety of ANNs and the human brain, despite their having different architectures, network initializations, learning objectives, and even different modalities such as language and vision [12,15–22]”
From The Umwelt Representation Hypothesis: rethinking Universality (2026) · cites this paper as [12]
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 2×
“These networks were 20 ResNet-18 architectures trained on image classification using the Tiny ImageNet dataset [11, 26, 19].”
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 1×
“Previous work has shown that in many neural networks trained on natural images, the first layer contains general-purpose V1-like filters tuned to orientation, frequency, and color, whereas subsequent layers contain filters that appear to be increasingly specialized [18, 31].”
Microsoft COCO: Common Objects in Context
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-10602-1_48 · Cited by 45,460
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380 · cited 2×
“All were trained on ImageNet for object classification [25] and obtained from the torchvision library [23, 28].”
PyTorch: An Imperative Style, High-Performance Deep Learning Library
Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. (2019) · arXiv · doi:10.48550/arxiv.1912.01703 · Cited by 31,473 · cited 3×
“All were trained on ImageNet for object classification [25] and obtained from the torchvision library [23, 28].”
Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification
He, K., Zhang, X., Ren, S., & Sun, J. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.123 · Cited by 19,741 · cited 1×
“The fourth set included 20 untrained ResNet-18 models [11] with different random weights, which were created using Kaiming normal initialization [12].”
Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
Reimers, N., & Gurevych, I. (2019) · Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) · doi:10.18653/v1/d19-1410 · Cited by 12,778
UMAP: Uniform Manifold Approximation and Projection
McInnes, L., Healy, J., Saul, N., & Großberger, L. (2018) · Journal of Open Source Software · doi:10.21105/joss.00861 · Cited by 9,826 · cited 1×
“We then selected the top 100 dimensions with the highest universality scores and visualized the image representations of these 100 dimensions in a 2D space using uniform manifold approximation and projection (UMAP) [21].”
Emergence of simple-cell receptive field properties by learning a sparse code for natural images
Olshausen, B. A., & Field, D. J. (1996) · Nature · doi:10.1038/381607a0 · Cited by 5,694 · cited 1×
“Theories of efficient coding suggest that frequency-and orientation-tuned filters are consistently observed in the first layer of vision systems because they constitute efficient bases that are adapted to the statistics of natural images [22, 27].”
Representational similarity analysis-connecting the branches of systems neuroscience
Kriegeskorte, N. (2008) · Frontiers in Systems Neuroscience · doi:10.3389/neuro.06.004.2008 · Cited by 3,355
Natural Image Statistics and Neural Representation
Simoncelli, E. P., & Olshausen, B. A. (2001) · Annual Review of Neuroscience · doi:10.1146/annurev.neuro.24.1.1193 · Cited by 2,460 · cited 1×
“Theories of efficient coding suggest that frequency-and orientation-tuned filters are consistently observed in the first layer of vision systems because they constitute efficient bases that are adapted to the statistics of natural images [22, 27].”
Performance-optimized hierarchical models predict neural responses in higher visual cortex
Yamins, D. L. K., Hong, H., Cadieu, C. F., Solomon, E. A., Seibert, D., & DiCarlo, J. J. (2014) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1403112111 · Cited by 1,924 · cited 1×
“Deep neural networks have a remarkable ability to simulate the representations of biological vision [15, 30, 5, 8].”
Using goal-driven deep learning models to understand sensory cortex
Yamins, D. L. K., & DiCarlo, J. J. (2016) · Nature Neuroscience · doi:10.1038/nn.4244 · Cited by 1,621 · cited 2×
“A leading approach interprets neural network representations in terms of their architectures and task objectives, which are thought to function as key constraints on a network’s learned representations [29, 24, 3, 7, 14].”
Understanding Neural Networks Through Deep Visualization
Yosinski, J., Clune, J., Nguyen, A. M., Fuchs, T. J., & Lipson, H. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1506.06579 · Cited by 1,358 · cited 1×
“Previous work has shown that in many neural networks trained on natural images, the first layer contains general-purpose V1-like filters tuned to orientation, frequency, and color, whereas subsequent layers contain filters that appear to be increasingly specialized [18, 31].”
Matching Categorical Object Representations in Inferior Temporal Cortex of Man and Monkey
Kriegeskorte, N., Mur, M., Ruff, D. A., Kiani, R., Bodurka, J., Esteky, H., Tanaka, K., & Bandettini, P. A. (2008) · Neuron · doi:10.1016/j.neuron.2008.10.043 · Cited by 1,344 · cited 1×
“Third, while previous work has revealed similarities between the visual cortex representations of humans and monkeys [17], we still know little about the degree to which representational dimensions may be universal or species-specific across mammalian vision.”
Kriegeskorte, N. (2015) · Annual Review of Vision Science · doi:10.1146/annurev-vision-082114-035447 · Cited by 1,093
A deep learning framework for neuroscience
Richards, B. A., Lillicrap, T. P., Beaudoin, P., Bengio, Y., Bogacz, R., Christensen, A., Clopath, C., Costa, R. P., de Berker, A., Ganguli, S., et al. (2019) · Nature Neuroscience · doi:10.1038/s41593-019-0520-2 · Cited by 1,027 · cited 2×
“A leading approach interprets neural network representations in terms of their architectures and task objectives, which are thought to function as key constraints on a network’s learned representations [29, 24, 3, 7, 14].”
Similarity of Neural Network Representations Revisited
Kornblith, S., Norouzi, M., Lee, H., & Hinton, G. (2019) · arXiv · doi:10.48550/arxiv.1905.00414 · Cited by 724
A Common, High-Dimensional Model of the Representational Space in Human Ventral Temporal Cortex
Haxby, J. V., Guntupalli, J. S., Connolly, A. C., Halchenko, Y. O., Conroy, B. R., Gobbini, M. I., Hanke, M., & Ramadge, P. J. (2011) · Neuron · doi:10.1016/j.neuron.2011.08.026 · Cited by 648
A massive 7T fMRI dataset to bridge cognitive neuroscience and artificial intelligence
Allen, E. J., St-Yves, G., Wu, Y., Breedlove, J. L., Prince, J. S., Dowdle, L. T., Nau, M., Caron, B., Pestilli, F., Charest, I., et al. (2022) · Nature Neuroscience · doi:10.1038/s41593-021-00962-x · Cited by 577 · cited 6×
“These regions are based on the “streams” masks as defined by the authors of the Natural Scenes Dataset study [1].”
InfluentialMPNet: Masked and Permuted Pre-training for Language Understanding
Song, K., Tan, X., Qin, T., Lu, J., & Liu, T. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2004.09297 · Cited by 514
Recurrence is required to capture the representational dynamics of the human visual system
Kietzmann, T. C., Spoerer, C. J., Sörensen, L. K. A., Cichy, R. M., Hauk, O., & Kriegeskorte, N. (2019) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1905544116 · Cited by 406
The neuroconnectionist research programme
Doerig, A., Sommers, R. P., Seeliger, K., Richards, B., Ismael, J., Lindsay, G. W., Kording, K. P., Konkle, T., van Gerven, M. A. J., Kriegeskorte, N., et al. (2023) · Nature Reviews Neuroscience · doi:10.1038/s41583-023-00705-w · Cited by 233 · cited 2×
“A leading approach interprets neural network representations in terms of their architectures and task objectives, which are thought to function as key constraints on a network’s learned representations [29, 24, 3, 7, 14].”
Using artificial neural networks to ask ‘why’ questions of minds and brains
Kanwisher, N., Khosla, M., & Dobs, K. (2023) · Trends in Neurosciences · doi:10.1016/j.tins.2022.12.008 · Cited by 182 · cited 2×
“A leading approach interprets neural network representations in terms of their architectures and task objectives, which are thought to function as key constraints on a network’s learned representations [29, 24, 3, 7, 14].”
High-level visual representations in the human brain are aligned with large language models
Doerig, A., Kietzmann, T. C., Allen, E., Wu, Y., Naselaris, T., Kay, K., & Charest, I. (2025) · Nature Machine Intelligence · doi:10.1038/s42256-025-01072-0 · Cited by 173
Conwell, C., Prince, J. S., Kay, K. N., Alvarez, G. A., & Konkle, T. (2022) · doi:10.1101/2022.03.28.485868 · Cited by 160 · cited 3×
“We computed conventional representational similarity analysis (RSA) scores for comparisons of networks and visual cortex [16], adapting the procedure described in [5].”
Intrinsic dimension of data representations in deep neural networks
Ansuini, A., Laio, A., Macke, J. H., & Zoccolan, D. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.12784 · Cited by 145
Garrido, Q., Balestriero, R., Najman, L. A., & LeCun, Y. (2022) · arXiv (Cornell University) · doi:10.48550/arxiv.2210.02885 · Cited by 135
A self-supervised domain-general learning framework for human ventral stream representation
Konkle, T., & Alvarez, G. A. (2022) · Nature Communications · doi:10.1038/s41467-022-28091-4 · Cited by 133
The Emergence of Semantic Meaning in the Ventral Temporal Pathway
Carlson, T. A., Simmons, R. A., Kriegeskorte, N., & Slevc, L. R. (2014) · Journal of Cognitive Neuroscience · doi:10.1162/jocn_a_00458 · Cited by 111 · cited 1×
“Previous work has shown that language-model embeddings of object names and scene captions are predictive of image representations in high-level visual cortex [4, 2, 6].”
Conwell, C., Prince, J. S., Kay, K. N., Alvarez, G. A., & Konkle, T. (2024) · Nature Communications · doi:10.1038/s41467-024-53147-y · Cited by 100
Cao, R., & Yamins, D. L. K. (2023) · Cognitive Systems Research · doi:10.1016/j.cogsys.2023.101200 · Cited by 97 · cited 2×
“A leading approach interprets neural network representations in terms of their architectures and task objectives, which are thought to function as key constraints on a network’s learned representations [29, 24, 3, 7, 14].”
Bonner, M. F., & Epstein, R. A. (2021) · Nature Communications · doi:10.1038/s41467-021-24368-2 · Cited by 96 · cited 1×
“Previous work has shown that language-model embeddings of object names and scene captions are predictive of image representations in high-level visual cortex [4, 2, 6].”
Self-citationSparse Autoencoders Find Highly Interpretable Features in Language Models
Cunningham, H., Ewart, A., Riggs, L., Huben, R., & Sharkey, L. (2023) · arXiv · doi:10.48550/arxiv.2309.08600 · Cited by 61
Unveiling functions of the visual cortex using task-specific deep neural networks
Dwivedi, K., Bonner, M. F., Cichy, R. M., & Roig, G. (2021) · PLOS Computational Biology · doi:10.1371/journal.pcbi.1009267 · Cited by 52
Self-citationHigh-performing neural network models of visual cortex benefit from high latent dimensionality
Elmoznino, E., & Bonner, M. F. (2024) · PLOS Computational Biology · doi:10.1371/journal.pcbi.1011792 · Cited by 49
Self-citationA Rainbow in Deep Network Black BoxesPreprint
Guth, F., Ménard, B., Rochette, G., & Mallat, S. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2305.18512 · Cited by 42
Model Zoos: A Dataset of Diverse Populations of Neural Network Models
Schürholt, K., Taskiran, D., Knyazev, B. A., Giró-i-Nieto, X., & Borth, D. (2022) · neural information processing systems · doi:10.52202/068431-2763 · Cited by 42
The Platonic Representation Hypothesis
Huh, M., Cheung, B., Wang, T., & Isola, P. (2024) · arXiv · doi:10.48550/arxiv.2405.07987 · Cited by 35 · cited 3×
“The invariance of these representations implies that they are not primarily governed by the details of a network’s architecture or task objective but instead by more general principles of natural image representation in deep vision systems [10, 13].”
Identifying Interpretable Visual Features in Artificial and Biological Neural SystemsPreprint
Klindt, D. A., Sanborn, S., Acosta, F., Poitevin, F., & Miolane, N. (2023) · Cold Spring Harbor Laboratory Institutional Repository (Cold Spring Harbor Laboratory) · doi:10.48550/arxiv.2310.11431 · Cited by 17
Flovik, V. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2405.01978 · Cited by 16
Universal scale-free representations in human visual cortex
Gauthaman, R. M., Ménard, B., & Bonner, M. (2025) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1013714 · Cited by 15
Self-citationOn the universality of neural encodings in CNNsPreprint
Guth, F., & Ménard, B. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2409.19460 · Cited by 5 · cited 2×
“The invariance of these representations implies that they are not primarily governed by the details of a network’s architecture or task objective but instead by more general principles of natural image representation in deep vision systems [10, 13].”
Model Zoos: A Dataset of Diverse Populations of Neural Network Models
Schürholt, K., Taskiran, D., Knyazev, B. A., et al. (2022) · Alexandria (UniSG) (University of St.Gallen) · doi:10.48550/arxiv.2209.14764 · Cited by 4 · cited 2×
“These networks were 20 ResNet-18 architectures trained on image classification using the Tiny ImageNet dataset [11, 26, 19].”
zche377/universal_dimensions: v1.0.0 - Initial Release
Chen, Z. (2025) · Zenodo (CERN European Organization for Nuclear Research) · doi:10.5281/zenodo.15347394 · Cited by 1
Self-citation
Research works printed but not linked: 4
- Y. Le, X. Yang, Tiny imagenet visual recognition challenge. CS 231N 7, 3 (2015).
- A. Krizhevsky I. Sutskever G. E. Hinton “ImageNet classification with deep convolutional neural networks ” in Advances in Neural Information Processing Systems F. Pereira C. Burges L. Bottou K. Weinberger Eds. (Curran Associates Inc. 2012) vol. 25 pp. 1097–1105.
- J. S. Prince G. Fajardo G. A. Alvarez T. Konkle “Manipulating dropout reveals an optimal balance of efficiency and robustness in biological and machine visual systems ” in The Twelfth International Conference on Learning Representations (Open Review 2024); https://openreview.net/pdf?id=ADDCErFzev.
- B. S. Robinson N. Drenkow C. Conwell M. Bonner “A sparse null code emerges in deep neural networks ” in Proceedings of UniReps: The First Workshop on Unifying Representations in Neural Models M. Fumero E. Rodolá C. Domine F. Locatello K. Dziugaite C. Mathilde Eds. (PMLR 2024) vol. 243 of Proceedings of Machine Learning Research pp. 302–314.
Cited by
The Universal Weight Subspace HypothesisPreprint
Kaushik, P., Chaudhari, S., Vaidya, A., Chellappa, R., & Yuille, A. (2025) · arXiv (Cornell University) · doi:10.48550/arxiv.2512.05117 · Cited by 37
Reanimating Images using Neural Representations of Dynamic Stimuli
Yeung, J., Luo, A. F., Sarch, G. H., Henderson, M. M., Ramanan, D., & Tarr, M. J. (2025) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52734.2025.00502 · Cited by 20
Convolutional architectures are cortex-aligned de novo
Kazemian, A., Elmoznino, E., & Bonner, M. (2025) · Nature Machine Intelligence · doi:10.1038/s42256-025-01142-3 · Cited by 24
“This suggests that for highly trained DNNs, architectural and objective-based constraints are relatively weak 8–10 .”
Self-citationUniversality of representation in biological and artificial neural networks
Hosseini, E., Casto, C., Zaslavsky, N., Conwell, C., Richardson, M., & Fedorenko, E. (2024) · doi:10.1101/2024.12.26.629294 · Cited by 10
Luo, A. F., Yeung, J., Zawar, R., Dewan, S., Henderson, M. M., Wehbe, L., & Tarr, M. J. (2024) · arXiv (Cornell University) · doi:10.48550/arxiv.2410.05266 · Cited by 16
“This observation raises an important question about the nature of each model’s learned features and their alignment with one another: Do these models converge upon similar feature representations (Chen & Bonner, 2024) for category selective brain regions despite their varied training paradigms?”
Background
Metadata
- Authors
- Chen, Z., & Bonner, M. F.
- Affiliations
- Johns Hopkins University (US)
- Corresponding author
- Chen, Zirui
- Venue
- Science Advances
- Published
- 2025-07-04
- Volume
- 11
- Issue
- 27
- Pages
- eadw7697-eadw7697
- ISSN
- 2375-2548 (ISSN-L)
- PMC
- PMC12219468
- OpenAlex
- W4411948297
- Cited by
- 16
- References
- 36
- Citations by year
- 2026: 12 · 2025: 3
- FWCI
- 8.51
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- Locations
- Science Advances (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by-nc-nd)
- OSF Preprints (OSF Preprints) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Life Sciences › Neuroscience › Cognitive Neuroscience › Visual perception and processing mechanisms
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- representational similarity, latent dimensions, universal representation, dimensionality reduction, functional MRI
- MeSH terms
- Brain, Brain Mapping, Humans, Image Processing, Computer-Assisted, Magnetic Resonance Imaging, Vision, Ocular, Visual Perception, Neural Networks, Computer
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.