pith. sign in

arxiv: 2605.20496 · v1 · pith:HR5X6TARnew · submitted 2026-05-19 · 🧬 q-bio.NC · cs.CV

Platonic Representations in the Human Brain: Unsupervised Recovery of Universal Geometry

Pablo Marcos-Manch\'on , Rishi Jha , Llu\'is Fuentemilla This is my paper

Pith reviewed 2026-05-21 06:04 UTC · model grok-4.3

classification 🧬 q-bio.NC cs.CV
keywords fMRIvisual cortexrepresentational geometryself-supervised learningcross-subject alignmentneural embeddingsorthogonal rotations
0
0 comments X

The pith

fMRI embeddings learned separately per person can be aligned across brains using only unsupervised orthogonal rotations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper trains a self-supervised encoder on each subject's fMRI responses to the same repeated visual scenes from the Natural Scenes Dataset, producing subject-specific embeddings without using any cross-subject data. It then shows that these independent spaces can be mapped onto one another by simple rotations that require no paired samples. When the rotations are further synchronized so all subjects share one common coordinate frame, cross-subject retrieval of the original stimuli improves. This pattern indicates that the learned representations preserve the underlying stimulus geometry in a way that is approximately the same across individuals.

Core claim

Independently learned subject-specific embeddings from fMRI can be translated across subjects using unsupervised orthogonal rotations, and synchronizing these rotations into a single shared latent space improves cross-subject retrieval, indicating that subject-specific fMRI representations are approximately isometric across individuals.

What carries the argument

Self-supervised encoder that learns subject-specific embeddings by exploiting repeated stimulus presentations within each individual.

If this is right

  • Subject-specific spaces are mutually compatible with a single common coordinate system.
  • Cross-subject retrieval performance rises once all pairwise rotations are synchronized to one shared space.
  • Purely geometric transformations suffice to translate between brains without paired cross-subject samples or intermediate models.
  • The results supply evidence that a shared neural geometry exists in human visual cortex.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the geometry is approximately universal, a learned rotation matrix could in principle predict one person's brain responses to a new scene from another's responses alone.
  • The same repeated-stimulus protocol could be applied to test whether comparable isometric structure appears in non-visual regions or during higher-level cognitive tasks.
  • Alignment quality might degrade systematically for subjects with atypical perceptual experience, offering a potential geometric signature of individual differences.

Load-bearing premise

The self-supervised encoder recovers a faithful embedding of the stimulus geometry rather than noise or subject-specific artifacts.

What would settle it

Orthogonal rotations between subjects produce no improvement in cross-subject stimulus retrieval accuracy compared with random or identity alignments on held-out trials.

Figures

Figures reproduced from arXiv: 2605.20496 by Llu\'is Fuentemilla, Pablo Marcos-Manch\'on, Rishi Jha.

Figure 1
Figure 1. Figure 1: Method overview. (A) Subject encoder. For each subject, fMRI responses are mapped into a low-dimensional embedding space using voxel reliability weighting, PCA, and multi-view CCA (MCCA), followed by a residual nonlinear refinement trained from repeated stimulus presentations. (B) Pairwise brain-to-brain translation. Independently learned subject embeddings are translated between subject pairs by estimatin… view at source ↗
Figure 2
Figure 2. Figure 2: Pairwise brain-to-brain translation. Performance for each ordered subject pair (s, t). Embeddings from source subject s are mapped into target subject t’s space using the unsupervised orthogonal transformation Rs→t and evaluated on the 515 held-out shared images. Mean Rank (average ± std: 2.56 ± 1.71, chance = 258) and R@1 (average ± std: 0.78 ± 0.14, chance = 0.002) measure image-level retrieval after tra… view at source ↗
Figure 3
Figure 3. Figure 3: Shared-space brain-to-brain translation. Retrieval performance after synchronizing pairwise brain-to-brain rotations into a single shared latent space. Each subject is mapped into the common coordinate system using one orthogonal transformation Rs, and retrieval is evaluated across ordered subject pairs on the 515 held-out shared images. Left: Mean Rank, lower is better (average ± std: 2.00 ± 0.76; chance … view at source ↗
read the original abstract

The Strong Platonic Representation Hypothesis suggests that representational convergence in artificial neural networks can be harnessed constructively: embeddings can be translated across models through a universal latent space without paired data. We ask whether an analogous geometry can be recovered across human brains. Using fMRI data from the Natural Scenes Dataset, we propose a self-supervised encoder that learns subject-specific embeddings from brain data alone by exploiting repeated stimulus presentations. We show that these independently learned spaces can be translated across subjects using unsupervised orthogonal rotations, without paired cross-subject samples or intermediate model representations. Synchronizing pairwise rotations into a single shared latent space further improves cross-subject retrieval, indicating that subject-specific spaces are mutually compatible with a common coordinate system. These results provide evidence for a shared neural geometry in the human visual cortex: subject-specific fMRI representations are approximately isometric across individuals and can be translated through purely geometric transformations.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes a self-supervised encoder to recover subject-specific embeddings from repeated fMRI presentations in the Natural Scenes Dataset. These independently learned spaces are shown to be alignable across subjects via unsupervised orthogonal rotations without paired cross-subject data; synchronizing the pairwise rotations into a single shared latent space further improves cross-subject retrieval. The results are interpreted as evidence that subject-specific fMRI representations are approximately isometric and can be translated through purely geometric transformations, supporting an analogue of the Strong Platonic Representation Hypothesis in human visual cortex.

Significance. If the central claim holds after appropriate controls, the work would provide novel evidence for a shared, stimulus-independent neural geometry across individuals that can be recovered in a fully unsupervised manner. This would strengthen analogies between biological and artificial representations and could inform subject-general decoding methods. The unsupervised, paired-data-free alignment and the use of repeated-stimulus self-supervision are methodologically attractive features.

major comments (2)
  1. Abstract and Results: the reported improvement in cross-subject retrieval after rotation synchronization is presented without quantitative metrics, error bars, statistical tests, or baseline comparisons, making it impossible to evaluate the effect size or rule out that the gain is driven by trivial shared stimulus-evoked variance rather than intrinsic geometry.
  2. Methods (self-supervised encoder section): because every subject views the identical Natural Scenes Dataset images, any encoder that extracts stimulus-driven signals will produce alignable spaces; the manuscript contains no control (e.g., comparison against a supervised decoder, correlation with independently measured stimulus distances, or a null model using only low-level visual features) that would falsify the alternative explanation that alignment reflects common response patterns rather than a universal Platonic geometry.
minor comments (1)
  1. Abstract: the phrase 'Strong Platonic Representation Hypothesis' is introduced without a concise definition or citation to the original ANN literature, which may confuse readers unfamiliar with the analogy.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript. We address each major comment below and outline the revisions we will make to strengthen the presentation and controls.

read point-by-point responses

  1. Referee: Abstract and Results: the reported improvement in cross-subject retrieval after rotation synchronization is presented without quantitative metrics, error bars, statistical tests, or baseline comparisons, making it impossible to evaluate the effect size or rule out that the gain is driven by trivial shared stimulus-evoked variance rather than intrinsic geometry.

    Authors: We agree that the main-text description of the synchronization improvement would benefit from explicit quantification. In the revised manuscript we will expand the Results section to report mean cross-subject retrieval accuracies (with standard errors across subjects), paired statistical tests comparing synchronized versus unsynchronized rotations, and baseline comparisons including random orthogonal matrices and a stimulus-average null model. These additions will allow direct evaluation of effect size and help address whether gains exceed those attributable to shared stimulus-evoked responses. revision: yes

  2. Referee: Methods (self-supervised encoder section): because every subject views the identical Natural Scenes Dataset images, any encoder that extracts stimulus-driven signals will produce alignable spaces; the manuscript contains no control (e.g., comparison against a supervised decoder, correlation with independently measured stimulus distances, or a null model using only low-level visual features) that would falsify the alternative explanation that alignment reflects common response patterns rather than a universal Platonic geometry.

    Authors: This concern is well-taken and highlights a possible confound. While the self-supervised objective uses repeat consistency to encourage stable representations, we will add the requested controls in the revised Methods and Results: (i) direct comparison of alignment performance against a supervised decoder trained on stimulus labels, (ii) correlation analyses between the learned embeddings and distances computed from low-level visual features (e.g., Gabor or pixel-based metrics), and (iii) a null model that permutes repeat identities during training. These will be presented alongside the main alignment results to help distinguish geometric compatibility from purely stimulus-driven commonality. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation remains self-contained

full rationale

The paper trains subject-specific self-supervised encoders independently on repeated stimulus presentations within each subject to produce embeddings, then applies post-hoc unsupervised orthogonal rotations to align those spaces and evaluates cross-subject retrieval. No equation or step reduces the claimed isometry or shared latent space to a definitional identity, fitted parameter renamed as prediction, or self-citation chain; the geometric alignment operates on independently derived representations without presupposing the target geometry in the training objective. The process is externally falsifiable via retrieval metrics and does not invoke load-bearing uniqueness theorems from the authors' prior work.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that repeated-stimulus self-supervision produces geometrically meaningful embeddings; no explicit free parameters or invented entities are named in the abstract, but the method implicitly assumes that orthogonal transformations suffice to align the spaces.

axioms (1)
  • domain assumption Repeated presentations of the same stimulus produce consistent enough brain responses to allow self-supervised embedding learning per subject.
    Invoked in the description of the encoder training on the Natural Scenes Dataset.

pith-pipeline@v0.9.0 · 5689 in / 1156 out tokens · 23307 ms · 2026-05-21T06:04:51.472047+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

66 extracted references · 66 canonical work pages · 1 internal anchor

  1. [1]

    Linguistic regularities in continuous space word representations

    Tomas Mikolov, Wen-tau Yih, and Geoffrey Zweig. Linguistic regularities in continuous space word representations. InProceedings of the 2013 Conference of the North American Chapter of the Associ- ation for Computational Linguistics: Human Language Technologies, pages 746–751. Association for Computational Linguistics, 2013

    work page 2013

  2. [2]

    Svcca: singular vector canonical correlation analysis for deep learning dynamics and interpretability

    Maithra Raghu, Justin Gilmer, Jason Yosinski, and Jascha Sohl-Dickstein. Svcca: singular vector canonical correlation analysis for deep learning dynamics and interpretability. InProceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 6078–6087, Red Hook, NY , USA,

    work page

  3. [3]

    ISBN 9781510860964

    Curran Associates Inc. ISBN 9781510860964

    work page

  4. [4]

    Similarity of neural network representations revisited

    Simon Kornblith, Mohammad Norouzi, Honglak Lee, and Geoffrey Hinton. Similarity of neural network representations revisited. InInternational Conference on Machine Learning (ICML), volume 97, pages 3519–3529, 2019

    work page 2019

  5. [5]

    Love, Christopher J Cueva, Erin Grant, Iris Groen, Jascha Achterberg, Joshua B

    Ilia Sucholutsky, Lukas Muttenthaler, Adrian Weller, Andi Peng, Andreea Bobu, Been Kim, Bradley C. Love, Christopher J Cueva, Erin Grant, Iris Groen, Jascha Achterberg, Joshua B. Tenenbaum, Kather- ine M. Collins, Katherine Hermann, Kerem Oktar, Klaus Greff, Martin N Hebart, Nathan Cloos, Nikolaus Kriegeskorte, Nori Jacoby, Qiuyi Zhang, Raja Marjieh, Robe...

    work page 2025

  6. [6]

    The platonic representation hypothesis

    Minyoung Huh, Brian Cheung, Tongzhou Wang, and Phillip Isola. The platonic representation hypothesis. InProceedings of the 41st International Conference on Machine Learning. JMLR.org, 2024

    work page 2024

  7. [7]

    Harnessing the universal geometry of embeddings

    Rishi Dev Jha, Collin Zhang, Vitaly Shmatikov, and John Xavier Morris. Harnessing the universal geometry of embeddings. InAdvances in Neural Information Processing Systems, 2025

    work page 2025

  8. [8]

    mini-vec2vec: Scaling universal geometry alignment with linear transformations, 2026

    Guy Dar. mini-vec2vec: Scaling universal geometry alignment with linear transformations, 2026. URL https://arxiv.org/abs/2510.02348

    work page arXiv 2026

  9. [9]

    Revisiting model stitching to compare neural repre- sentations

    Yamini Bansal, Preetum Nakkiran, and Boaz Barak. Revisiting model stitching to compare neural repre- sentations. In A. Beygelzimer, Y . Dauphin, P. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Processing Systems, 2021. URLhttps://openreview.net/forum?id=ak06J5jNR4

    work page 2021

  10. [10]

    Representation potentials of foundation models for multimodal alignment: A survey

    Jianglin Lu, Hailing Wang, Yi Xu, Yizhou Wang, Kuo Yang, and Yun Fu. Representation potentials of foundation models for multimodal alignment: A survey. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 16669–16684. Association for Computational Linguistics, 2025. ISBN 979-8-89176-332-6. doi: 10.18653/v1/2025.e...

    work page doi:10.18653/v1/2025.emnlp-main.843 2025

  11. [11]

    Intersubject synchronization of cortical activity during natural vision.Science, 303(5664):1634–1640, 2004

    Uri Hasson, Yuval Nir, Ifat Levy, Galit Fuhrmann, and Rafael Malach. Intersubject synchronization of cortical activity during natural vision.Science, 303(5664):1634–1640, 2004. doi: 10.1126/science.1089506

    work page doi:10.1126/science.1089506 2004

  12. [12]

    Measuring shared responses across subjects using intersubject correlation.Social Cognitive and Affective Neuroscience, 14(6):667–685,

    Samuel A Nastase, Valeria Gazzola, Uri Hasson, and Christian Keysers. Measuring shared responses across subjects using intersubject correlation.Social Cognitive and Affective Neuroscience, 14(6):667–685,

    work page

  13. [13]

    doi: 10.1093/scan/nsz037

    work page doi:10.1093/scan/nsz037

  14. [14]

    Representational similarity analysis - connecting the branches of systems neuroscience,

    Nikolaus Kriegeskorte, Marieke Mur, and Peter A. Bandettini. Representational similarity analysis – connecting the branches of systems neuroscience.Frontiers in Systems Neuroscience, 2, 2008. doi: 10.3389/neuro.06.004.2008

    work page doi:10.3389/neuro.06.004.2008 2008

  15. [15]

    The topology and geometry of neural representations.Proceedings of the National Academy of Sciences, 121(42):e2317881121, 2024

    Baihan Lin and Nikolaus Kriegeskorte. The topology and geometry of neural representations.Proceedings of the National Academy of Sciences, 121(42):e2317881121, 2024. doi: 10.1073/pnas.2317881121. URL https://www.pnas.org/doi/abs/10.1073/pnas.2317881121. 10

    work page doi:10.1073/pnas.2317881121 2024

  16. [16]

    Saarinen, M.-L

    James V . Haxby, J. Swaroop Guntupalli, Andrew C. Connolly, Yaroslav O. Halchenko, Bryan R. Conroy, M. Ida Gobbini, Michael Hanke, and Peter J. Ramadge. A common, high-dimensional model of the representational space in human ventral temporal cortex.Neuron, 72(2):404–416, 2011. doi: 10.1016/j. neuron.2011.08.026

    work page doi:10.1016/j 2011

  17. [17]

    Swaroop Guntupalli, Michael Hanke, Yaroslav O

    J. Swaroop Guntupalli, Michael Hanke, Yaroslav O. Halchenko, Andrew C. Connolly, Peter J. Ramadge, and James V . Haxby. A model of representational spaces in human cortex.Cerebral Cortex, 26(6): 2919–2934, 2016. doi: 10.1093/cercor/bhw068

    work page doi:10.1093/cercor/bhw068 2016

  18. [18]

    CLIP-MUSED: CLIP-guided multi- subject visual neural information semantic decoding

    Qiongyi Zhou, Changde Du, Shengpei Wang, and Huiguang He. CLIP-MUSED: CLIP-guided multi- subject visual neural information semantic decoding. InThe Twelfth International Conference on Learning Representations, 2024. URLhttps://openreview.net/forum?id=lKxL5zkssv

    work page 2024

  19. [19]

    Functional brain-to-brain transformation without shared stimuli.NeuroImage, 327:121741, 2026

    Navve Wasserman, Roman Beliy, Roy Urbach, and Michal Irani. Functional brain-to-brain transformation without shared stimuli.NeuroImage, 327:121741, 2026. doi: 10.1016/j.neuroimage.2026.121741

    work page doi:10.1016/j.neuroimage.2026.121741 2026

  20. [20]

    Allen, Ghislain St-Yves, Yihan Wu, Jesse L

    Emily J. Allen, Ghislain St-Yves, Yihan Wu, Jesse L. Breedlove, Jacob S. Prince, Logan T. Dowdle, Matthias Nau, Brad Caron, Franco Pestilli, Ian Charest, J. Benjamin Hutchinson, Thomas Naselaris, and Kendrick Kay. A massive 7t fMRI dataset to bridge cognitive neuroscience and artificial intelligence. Nature Neuroscience, 25(1):116–126, 2022. doi: 10.1038/...

    work page doi:10.1038/s41593-021-00962-x 2022

  21. [21]

    Thomas T. Liu. Noise contributions to the fmri signal: An overview.NeuroImage, 143:141–151, 2016. doi: 10.1016/j.neuroimage.2016.09.008

    work page doi:10.1016/j.neuroimage.2016.09.008 2016

  22. [22]

    Improving the accuracy of single-trial fmri response estimates using glmsingle.eLife, 11:e77599, nov

    Jacob S Prince, Ian Charest, Jan W Kurzawski, John A Pyles, Michael J Tarr, and Kendrick N Kay. Improving the accuracy of single-trial fmri response estimates using glmsingle.eLife, 11:e77599, nov

    work page

  23. [24]

    Natural scene reconstruction from fmri signals using generative latent diffusion.Scientific Reports, 13(1):15666, Sep 2023

    Furkan Ozcelik and Rufin VanRullen. Natural scene reconstruction from fmri signals using generative latent diffusion.Scientific Reports, 13(1):15666, Sep 2023. doi: 10.1038/s41598-023-42891-8

    work page doi:10.1038/s41598-023-42891-8 2023

  24. [25]

    Reconstructing the mind’s eye: fMRI-to-image with contrastive learning and diffusion priors

    Paul Steven Scotti, Atmadeep Banerjee, Jimmie Goode, Stepan Shabalin, Alex Nguyen, Cohen Ethan, Aidan James Dempster, Nathalie Verlinde, Elad Yundler, David Weisberg, Kenneth Norman, and Tan- ishq Mathew Abraham. Reconstructing the mind’s eye: fMRI-to-image with contrastive learning and diffusion priors. InThirty-seventh Conference on Neural Information P...

    work page 2023

  25. [26]

    J. R. KETTENRING. Canonical analysis of several sets of variables.Biometrika, 58(3):433–451, 1971. doi: 10.1093/biomet/58.3.433

    work page doi:10.1093/biomet/58.3.433 1971

  26. [27]

    Regularized generalized canonical correlation analysis.Psy- chometrika, 76(2):257–284, 2011

    Arthur Tenenhaus and Michel Tenenhaus. Regularized generalized canonical correlation analysis.Psy- chometrika, 76(2):257–284, 2011

    work page 2011

  27. [28]

    Representation Learning with Contrastive Predictive Coding

    Aaron van den Oord, Yazhe Li, and Oriol Vinyals. Representation learning with contrastive predictive coding, 2019. URLhttps://arxiv.org/abs/1807.03748

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  28. [29]

    Besl and Neil D

    Paul J. Besl and Neil D. McKay. A method for registration of 3-d shapes.IEEE Trans. Pattern Anal. Mach. Intell., 14(2), 1992. doi: 10.1109/34.121791

    work page doi:10.1109/34.121791 1992

  29. [30]

    A. Singer. Angular synchronization by eigenvectors and semidefinite programming.Applied and Compu- tational Harmonic Analysis, 30(1):20–36, 2011. doi: 10.1016/j.acha.2010.02.001

    work page doi:10.1016/j.acha.2010.02.001 2011

  30. [31]

    Exact and stable recovery of rotations for robust synchronization.Informa- tion and Inference: A Journal of the IMA, 2:145–193, 10 2013

    Lanhui Wang and Amit Singer. Exact and stable recovery of rotations for robust synchronization.Informa- tion and Inference: A Journal of the IMA, 2:145–193, 10 2013. doi: 10.1093/imaiai/iat005

    work page doi:10.1093/imaiai/iat005 2013

  31. [32]

    Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C

    Tsung-Yi Lin, Michael Maire, Serge J. Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C. Lawrence Zitnick. Microsoft COCO: Common objects in context. InEuropean Conference on Computer Vision (ECCV), pages 740–755, 2014

    work page 2014

  32. [33]

    Prince, Ian Charest, Jan W

    Jacob S. Prince, Ian Charest, Jan W. Kurzawski, John A. Pyles, Michael J. Tarr, and Kendrick N. Kay. Improving the accuracy of single-trial fMRI response estimates using GLMsingle.eLife, 11, 2022. doi: 10.7554/eLife.77599

    work page doi:10.7554/elife.77599 2022

  33. [34]

    On deep multi-view representation learning

    Weiran Wang, Raman Arora, Karen Livescu, and Jeff Bilmes. On deep multi-view representation learning. InProceedings of the 32nd International Conference on Machine Learning, volume 37, pages 1083–1092,

    work page

  34. [35]

    URLhttps://proceedings.mlr.press/v37/wangb15.html

    work page

  35. [36]

    Deep canonical correlation analysis

    Galen Andrew, Raman Arora, Jeff Bilmes, and Karen Livescu. Deep canonical correlation analysis. In Proceedings of the 30th International Conference on Machine Learning, volume 28, pages 1247–1255,

    work page

  36. [37]

    URLhttps://proceedings.mlr.press/v28/andrew13.html. 11

    work page

  37. [38]

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mido Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Herve Jegou, Julien Mairal, Patrick L...

    work page 2024

  38. [39]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision. InInternational Conference on Machine Learning (ICML), pages 8748–8763, 2021

    work page 2021

  39. [40]

    How to train your ViT? data, augmentation, and regularization in vision transformers.Transactions on Machine Learning Research, 2022

    Andreas Peter Steiner, Alexander Kolesnikov, Xiaohua Zhai, Ross Wightman, Jakob Uszkoreit, and Lucas Beyer. How to train your ViT? data, augmentation, and regularization in vision transformers.Transactions on Machine Learning Research, 2022. URLhttps://openreview.net/forum?id=4nPswr1KcP

    work page 2022

  40. [41]

    Sentence-BERT: Sentence embeddings using Siamese BERT-networks

    Nils Reimers and Iryna Gurevych. Sentence-BERT: Sentence embeddings using Siamese BERT-networks. InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992,

    work page 2019

  41. [42]

    doi: 10.18653/v1/D19-1410

    work page doi:10.18653/v1/d19-1410

  42. [43]

    Yixuan Li, Jason Yosinski, Jeff Clune, Hod Lipson, and John Hopcroft. Convergent learning: Do different neural networks learn the same representations? InProceedings of the 1st International Workshop on Feature Extraction: Modern Questions and Challenges at NIPS 2015, volume 44 ofProceedings of Machine Learning Research, pages 196–212. PMLR, 2015

    work page 2015

  43. [44]

    Gromov-Wasserstein alignment of word embedding spaces

    David Alvarez-Melis and Tommi Jaakkola. Gromov-Wasserstein alignment of word embedding spaces. InProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1881–1890. Association for Computational Linguistics, 2018. doi: 10.18653/v1/D18-1214

    work page doi:10.18653/v1/d18-1214 2018

  44. [45]

    Unsupervised alignment of embeddings with wasserstein procrustes

    Edouard Grave, Armand Joulin, and Quentin Berthet. Unsupervised alignment of embeddings with wasserstein procrustes. InProceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, volume 89 ofProceedings of Machine Learning Research, pages 1880–1890, 2019

    work page 2019

  45. [46]

    Relative representations enable zero-shot latent space communication

    Luca Moschella, Valentino Maiorca, Marco Fumero, Antonio Norelli, Francesco Locatello, and Emanuele Rodolà. Relative representations enable zero-shot latent space communication. InThe Eleventh Interna- tional Conference on Learning Representations, 2023. URL https://openreview.net/forum?id= SrC-nwieGJ

    work page 2023

  46. [47]

    Word translation without parallel data

    Guillaume Lample, Alexis Conneau, Marc’Aurelio Ranzato, Ludovic Denoyer, and Hervé Jégou. Word translation without parallel data. InInternational Conference on Learning Representations, 2018. URL https://openreview.net/forum?id=H196sainb

    work page 2018

  47. [48]

    Shared representations in brains and models reveal a two-route cortical organization during scene perception.Communications Biology, May 2026

    Pablo Marcos-Manchón and Lluís Fuentemilla. Shared representations in brains and models reveal a two-route cortical organization during scene perception.Communications Biology, May 2026. doi: 10.1038/s42003-026-10169-0

    work page doi:10.1038/s42003-026-10169-0 2026

  48. [49]

    Honey, Chung H

    Janice Chen, Yuan Chang Leong, Christopher J. Honey, Chung H. Yong, Kenneth A. Norman, and Uri Hasson. Shared memories reveal shared structure in neural activity across individuals.Nature Neuroscience, 20(1):115–125, 2017. doi: 10.1038/nn.4450

    work page doi:10.1038/nn.4450 2017

  49. [50]

    Representational models: A common framework for understanding encoding, pattern-component, and representational-similarity analysis.PLOS Computational Biology, 13(4):1–33, 2017

    Jörn Diedrichsen and Nikolaus Kriegeskorte. Representational models: A common framework for understanding encoding, pattern-component, and representational-similarity analysis.PLOS Computational Biology, 13(4):1–33, 2017. doi: 10.1371/journal.pcbi.1005508

    work page doi:10.1371/journal.pcbi.1005508 2017

  50. [51]

    Unsupervised method for representation transfer from one brain to another.Frontiers in Neuroinformatics, V olume 18 - 2024, 2024

    Daiki Nakamura, Shizuo Kaji, Ryota Kanai, and Ryusuke Hayashi. Unsupervised method for representation transfer from one brain to another.Frontiers in Neuroinformatics, V olume 18 - 2024, 2024. doi: 10.3389/ fninf.2024.1470845

    work page arXiv 2024

  51. [52]

    Bazeille, H

    T. Bazeille, H. Richard, H. Janati, and B. Thirion. Local optimal transport for functional brain template estimation. In Albert C. S. Chung, James C. Gee, Paul A. Yushkevich, and Siqi Bao, editors,Information Processing in Medical Imaging, pages 237–248. Springer International Publishing, 2019

    work page 2019

  52. [53]

    An empirical evaluation of functional alignment using inter-subject decoding.NeuroImage, 245:118683, 2021

    Thomas Bazeille, Elizabeth DuPre, Hugo Richard, Jean-Baptiste Poline, and Bertrand Thirion. An empirical evaluation of functional alignment using inter-subject decoding.NeuroImage, 245:118683, 2021. doi: 10.1016/j.neuroimage.2021.118683. 12

    work page doi:10.1016/j.neuroimage.2021.118683 2021

  53. [54]

    Daniel L. K. Yamins, Ha Hong, Charles F. Cadieu, Ethan A. Solomon, Darren Seibert, and James J. DiCarlo. Performance-optimized hierarchical models predict neural responses in higher visual cortex.Proceedings of the National Academy of Sciences, 111(23):8619–8624, 2014. doi: 10.1073/pnas.1403112111

    work page doi:10.1073/pnas.1403112111 2014

  54. [55]

    Deep supervised, but not unsupervised, models may explain IT cortical representation.PLOS Computational Biology, 10(11):1–29, 2014

    Seyed-Mahdi Khaligh-Razavi and Nikolaus Kriegeskorte. Deep supervised, but not unsupervised, models may explain IT cortical representation.PLOS Computational Biology, 10(11):1–29, 2014. doi: 10.1371/ journal.pcbi.1003915

    work page 2014

  55. [56]

    Comparison of deep neural networks to spatio-temporal cortical dynamics of human visual object recognition reveals hierarchical correspondence.Scientific Reports, 6(1):27755, 2016

    Radoslaw Martin Cichy, Aditya Khosla, Dimitrios Pantazis, Antonio Torralba, and Aude Oliva. Comparison of deep neural networks to spatio-temporal cortical dynamics of human visual object recognition reveals hierarchical correspondence.Scientific Reports, 6(1):27755, 2016. doi: 10.1038/srep27755

    work page doi:10.1038/srep27755 2016

  56. [57]

    Martin Schrimpf, Jonas Kubilius, Michael J. Lee, N. Apurva Ratan Murty, Robert Ajemian, and James J. DiCarlo. Integrative benchmarking to advance neurally mechanistic models of human intelligence.Neuron, 108(3):413–423, 2020. doi: 10.1016/j.neuron.2020.07.040

    work page doi:10.1016/j.neuron.2020.07.040 2020

  57. [58]

    Brains and algorithms partially converge in natural language processing.Communications Biology, 5(1):134, 2022

    Charlotte Caucheteux and Jean-Rémi King. Brains and algorithms partially converge in natural language processing.Communications Biology, 5(1):134, 2022. doi: 10.1038/s42003-022-03036-1

    work page doi:10.1038/s42003-022-03036-1 2022

  58. [59]

    Talia Konkle and George A. Alvarez. A self-supervised domain-general learning framework for human ven- tral stream representation.Nature Communications, 13(1):491, 2022. doi: 10.1038/s41467-022-28091-4

    work page doi:10.1038/s41467-022-28091-4 2022

  59. [60]

    Wang, Kendrick Kay, Thomas Naselaris, Michael J

    Aria Y . Wang, Kendrick Kay, Thomas Naselaris, Michael J. Tarr, and Leila Wehbe. Better models of human high-level visual cortex emerge from natural language supervision with a large and diverse dataset. Nature Machine Intelligence, 5(12):1415–1426, 2023. doi: 10.1038/s42256-023-00753-y

    work page doi:10.1038/s42256-023-00753-y 2023

  60. [61]

    Kietzmann, Emily Allen, Yihan Wu, Thomas Naselaris, Kendrick Kay, and Ian Charest

    Adrien Doerig, Tim C. Kietzmann, Emily Allen, Yihan Wu, Thomas Naselaris, Kendrick Kay, and Ian Charest. High-level visual representations in the human brain are aligned with large language models. Nature Machine Intelligence, pages 1220–1234, 2025. doi: 10.1038/s42256-025-01072-0

    work page doi:10.1038/s42256-025-01072-0 2025

  61. [62]

    Prince, Kendrick N

    Colin Conwell, Jacob S. Prince, Kendrick N. Kay, George A. Alvarez, and Talia Konkle. A large-scale examination of inductive biases shaping high-level visual representation in brains and machines.Nature Communications, 15(1):9383, 2024. doi: 10.1038/s41467-024-53147-y

    work page doi:10.1038/s41467-024-53147-y 2024

  62. [63]

    Zirui Chen and Michael F. Bonner. Universal dimensions of visual representation.Science Advances, 11 (27):eadw7697, 2025. doi: 10.1126/sciadv.adw7697

    work page doi:10.1126/sciadv.adw7697 2025

  63. [64]

    Glasser, Timothy S

    Matthew F. Glasser, Timothy S. Coalson, Emma C. Robinson, Carl D. Hacker, John Harwell, Essa Yacoub, Kamil Ugurbil, Jesper Andersson, Christian F. Beckmann, Mark Jenkinson, Stephen M. Smith, and David C. Van Essen. A multi-modal parcellation of human cerebral cortex.Nature, 536(7615):171–178, 2016. doi: 10.1038/nature18933

    work page doi:10.1038/nature18933 2016

  64. [65]

    Mruczek, Michael J

    Liang Wang, Ryan E.B. Mruczek, Michael J. Arcaro, and Sabine Kastner. Probabilistic maps of visual topography in human cortex.Cerebral Cortex, 25(10):3911–3931, 2015. doi: 10.1093/cercor/bhu277

    work page doi:10.1093/cercor/bhu277 2015

  65. [66]

    Pytorch image models, 2019

    Ross Wightman. Pytorch image models, 2019

    work page 2019

  66. [67]

    Neural rays for occlusion-aware image-based rendering,

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners. InIEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15979–15988, 2022. doi: 10.1109/CVPR52688.2022.01553. 13 Appendix A fMRI preprocessing details For all main analyses, we use the Natural Scenes D...

    work page doi:10.1109/cvpr52688.2022.01553 2022