pith. sign in

arxiv: 2605.28990 · v1 · pith:WH472CPGnew · submitted 2026-05-27 · 💻 cs.LG

Learning Robust and Task-Invariant Functional Representation from fMRI through Siamese Self-Supervised Learning

Jiyao Wang , Peiyu Duan , Nicha C. Dvornek , Lawrence H. Staib , Denis Sukhodolsky , Pamela Ventola , James S. Duncan This is my paper

Pith reviewed 2026-06-29 13:29 UTC · model grok-4.3

classification 💻 cs.LG
keywords fMRIself-supervised learningSiamese networksbrain representationtask-invariant featuresneuroimagingdata-efficient learningpositive pairs
0
0 comments X

The pith

Self-supervised learning from positive-only fMRI pairs produces task-general representations that beat supervised baselines.

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

The paper presents BrainSimSiam, a lightweight Siamese framework that trains on positive-only pairs drawn from fMRI scans to extract features. These features are shown to support multiple downstream classification and regression tasks on neurological data while requiring far less compute than large foundation models. The approach targets the common constraints of small samples, variable labels, and high dimensionality in neuroimaging. A reader would care because it offers a practical route to usable brain representations when labeled data and pretraining budgets are limited.

Core claim

BrainSimSiam leverages positive-only data pairs to learn robust and generalizable features from fMRI, achieving strong performance across multiple downstream classification and regression tasks, outperforming fully supervised baselines and approaching the performance of large-scale models.

What carries the argument

BrainSimSiam, a Siamese self-supervised network that learns task-invariant representations solely from positive pairs without negative samples or labels.

If this is right

  • Representations can be fine-tuned for new classification tasks on psychiatric conditions with only small labeled sets.
  • Regression on continuous brain-function measures becomes feasible without task-specific pretraining.
  • Research groups with limited compute can reach performance levels previously requiring large-scale pretraining.
  • The same positive-pair approach may reduce reliance on combining multiple datasets for foundation-model training.

Where Pith is reading between the lines

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

  • The method could extend to other high-dimensional time-series signals such as EEG if the positive-pair structure transfers.
  • If the representations prove scanner-invariant, they might enable pooling across sites without explicit harmonization steps.
  • Testing whether the features remain stable when the positive pairs come from different acquisition protocols would clarify robustness limits.

Load-bearing premise

Positive-only pairs drawn from fMRI scans contain sufficient structure to learn task-invariant features that generalize without being dominated by noise, scanner effects, or dataset-specific artifacts.

What would settle it

Train BrainSimSiam on one fMRI collection and evaluate on a completely held-out multi-site dataset for a new psychiatric condition; failure to exceed a supervised model trained on the target data would falsify the generalization claim.

Figures

Figures reproduced from arXiv: 2605.28990 by Denis Sukhodolsky, James S. Duncan, Jiyao Wang, Lawrence H. Staib, Nicha C. Dvornek, Pamela Ventola, Peiyu Duan.

Figure 1
Figure 1. Figure 1: Example of biological and scrambled motion tasks in the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Model architectures. (a) General SimSiam framework. (b) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Pipeline of the BrainSimSiam pretraining and downstream application [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Illustrations of augmentation operators used in training and interpretation. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Heatmaps of maximum absolute correlation of any single feature channels of learned embedding with metrics. [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Pearson’s correlation between tested scores [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Plots of quantitative performance on the classification and regression tasks from HCP and Biopoint dataset. Three training settings [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Heatmaps of node importance in HCP and Biopoint. Darker regions indicate higher importance for generating embedding or [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
read the original abstract

Functional magnetic resonance imaging (fMRI) is a powerful tool for investigating human brain function. However, the high cost of data acquisition and the inherent subjectivity of psychiatric rating scales often lead to datasets with small sample sizes and variable label quality, especially when targeting a specific neurological condition. Combined with the inherently high dimensionality of fMRI data, these limitations substantially increase the risk of model overfitting. Recent years have seen growing interest in developing fMRI foundation models by combining multiple datasets; however, the computational resources needed for pretraining and fine-tuning are often prohibitive. We show that a lightweight self-supervised framework yields representations that generalize across diverse downstream tasks, outperforming fully supervised baselines and approaching the performance of large-scale models. We introduce BrainSimSiam, a data-efficient self-supervised representation learning framework that leverages positive-only data pairs to learn robust and generalizable features. We demonstrate that the learned representations achieve strong performance across multiple downstream classification and regression tasks, highlighting the potential of BrainSimSiam for data-limited neuroimaging applications.

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

1 major / 0 minor

Summary. The paper introduces BrainSimSiam, a lightweight Siamese self-supervised learning framework that uses positive-only pairs from fMRI scans to learn robust, task-invariant functional representations. It claims these representations generalize across diverse downstream classification and regression tasks, outperforming fully supervised baselines while approaching the performance of large-scale foundation models, and are particularly suited to data-limited neuroimaging settings with small samples and noisy labels.

Significance. If the empirical results hold, the contribution would be significant for data-efficient fMRI representation learning: it offers a computationally lightweight alternative to large-scale pretraining that could reduce overfitting risks in psychiatric neuroimaging without requiring extensive labeled data or prohibitive resources.

major comments (1)
  1. Abstract: the central claim that the learned representations 'outperform fully supervised baselines and approach the performance of large-scale models' on classification and regression tasks is asserted without any quantitative results, error bars, dataset sizes, baseline implementations, or ablation evidence. This prevents evaluation of whether the positive-only pair construction actually yields task-invariant features that generalize beyond scanner or dataset artifacts.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their thoughtful review and for highlighting the need for stronger substantiation in the abstract. We address the major comment below.

read point-by-point responses

  1. Referee: [—] Abstract: the central claim that the learned representations 'outperform fully supervised baselines and approach the performance of large-scale models' on classification and regression tasks is asserted without any quantitative results, error bars, dataset sizes, baseline implementations, or ablation evidence. This prevents evaluation of whether the positive-only pair construction actually yields task-invariant features that generalize beyond scanner or dataset artifacts.

    Authors: We agree that the abstract, as currently written, asserts performance claims without accompanying quantitative details. The full manuscript contains the requested elements: quantitative results with error bars across multiple datasets and tasks, descriptions of dataset sizes and preprocessing, implementation details for baselines (including supervised models and large-scale foundation models), and ablations on the positive-only pair construction. These results are presented in Sections 4 and 5 with statistical comparisons. To address the concern directly, we will revise the abstract to include key quantitative metrics (e.g., mean accuracy or correlation values with standard deviations) and a brief note on the evaluation setup. Regarding generalization beyond scanner or dataset artifacts, the experiments include cross-dataset and cross-scanner evaluations that support task-invariance; we will ensure the revised abstract references this evidence more explicitly. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces BrainSimSiam, a Siamese self-supervised framework for fMRI representations using positive-only pairs. No equations, derivations, or parameter-fitting steps are described that reduce claimed performance or task-invariance to inputs by construction. The central claims rest on empirical generalization across downstream tasks rather than any self-definitional, fitted-input, or self-citation load-bearing logic. The provided abstract and method summary contain no quoted reductions matching the enumerated circularity patterns, making the result self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are stated or derivable from the provided text.

pith-pipeline@v0.9.1-grok · 5735 in / 1004 out tokens · 50941 ms · 2026-06-29T13:29:12.299979+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

48 extracted references · 39 canonical work pages · 6 internal anchors

  1. [1]

    Abuhantash, M

    F. Abuhantash, M. Abuhantash, A. AlShehhi, Comorbidity-based framework for alzheimer’s disease classification using graph neural networks, Scientific Reports (09 2024). doi:10.1038/s41598-024-72321-2

    work page doi:10.1038/s41598-024-72321-2 2024

  2. [2]

    L. Chen, Y. Yang, A. Yu, S. Guo, K. Ren, Q. Liu, C. Qiao, An explainable spatio-temporal graph convolutional network for the biomark- ers identification of ADHD, Biomedical Sig- nal Processing and Control 99 (2025) 106913. doi:https://doi.org/10.1016/j.bspc.2024.106913

    work page doi:10.1016/j.bspc.2024.106913 2025

  3. [3]

    X. Li, Y. Zhou, N. Dvornek, M. Zhang, S. Gao, J. Zhuang, D. Scheinost, L. H. Staib, P. Ven- tola, J. S. Duncan, BrainGNN: Interpretable brain graph neural network for fmri analy- sis, Medical Image Analysis 74 (2021) 102233. doi:https://doi.org/10.1016/j.media.2021.102233

    work page doi:10.1016/j.media.2021.102233 2021

  4. [4]

    Ferreira, A

    D. Ferreira, A. Nordberg, E. West- man, Biological subtypes of alzheimer dis- ease, Neurology 94 (10) (2020) 436–448. doi:10.1212/WNL.0000000000009058

    work page doi:10.1212/wnl.0000000000009058 2020

  5. [5]

    H. M. Geurts, S. Verté, J. Oosterlaan, H. Roey- ers, J. A. Sergeant, ADHD subtypes: do they dif- fer in their executive functioning profile?, Archives of Clinical Neuropsychology 20 (4) (2005) 457–477. doi:10.1016/j.acn.2004.11.001

    work page doi:10.1016/j.acn.2004.11.001 2005

  6. [6]

    Craig, A

    F. Craig, A. Crippa, M. Ruggiero, V. Rizzato, L. Russo, I. Fanizza, A. Trabacca, Charac- terization of autism spectrum disorder (asd) subtypes based on the relationship between mo- tor skills and social communication abilities, Human Movement Science 77 (2021) 102802. doi:https://doi.org/10.1016/j.humov.2021.102802

    work page doi:10.1016/j.humov.2021.102802 2021

  7. [7]

    H. Song, M. Kim, D. Park, Y. Shin, J.-G. Lee, Learning from noisy labels with deep neural net- works: A survey, IEEE Transactions on Neural Net- works and Learning Systems 34 (11) (2023) 8135–

    2023

  8. [8]

    doi:10.1109/TNNLS.2022.3152527

    work page doi:10.1109/tnnls.2022.3152527 2022

  9. [9]

    N. C. Dvornek, D. Yang, P. Ventola, J. S. Dun- can, Learning generalizable recurrent neural networks from small task-fmri datasets, in: International Con- ference on Medical Image Computing and Computer- Assisted Intervention, Springer, 2018, pp. 329–337

    2018

  10. [10]

    J. Wang, N. C. Dvornek, P. Duan, L. H. Staib, P. Ven- tola, J. S. Duncan, STNAGNN: Data-driven spatio- temporal brain connectivity beyond FC, in: Medical Imaging with Deep Learning, 2025

    2025

  11. [11]

    G. Shi, Y. Yao, Y. Zhu, X. Lin, L. Ji, W. Liu, X. Li, Contrastive hierarchical augmentation learning for modeling cognitive and multimodal brain network, IEEE Transactions on Computational Social Systems (2024) 1–11doi:10.1109/TCSS.2024.3402328

    work page doi:10.1109/tcss.2024.3402328 2024

  12. [12]

    X. Wang, L. Yao, I. Rekik, Y. Zhang, Contrastive functional connectivity graph learning for population- based fmri classification, in: L. Wang, Q. Dou, P. T. Fletcher, S.Speidel, S.Li(Eds.), MedicalImageCom- puting and Computer Assisted Intervention – MIC- CAI 2022, Springer Nature Switzerland, Cham, 2022, pp. 221–230

    2022

  13. [13]

    S. I. Ktena, S. Parisot, E. Ferrante, M. Rajchl, M. Lee, B. Glocker, D. Rueckert, Metric learning with spectral graph convolutions on brain connec- tivity networks, NeuroImage 169 (2018) 431–442. doi:https://doi.org/10.1016/j.neuroimage.2017.12.052

    work page doi:10.1016/j.neuroimage.2017.12.052 2018

  14. [14]

    X. Wang, Y. Chu, Q. Wang, L. Cao, L. Qiao, L. Zhang, M. Liu, Unsupervised contrastive graph learning for resting-state functional mri analysis and brain disorder detection, Hu- man Brain Mapping 44 (17) (2023) 5672–5692. doi:https://doi.org/10.1002/hbm.26469

    work page doi:10.1002/hbm.26469 2023

  15. [15]

    Y. Zhou, P. Duan, Y. Du, N. C. Dvornek, Self- supervised pre-training tasks for an fmri time-series transformer in autism detection, in: International Workshop on Machine Learning in Clinical Neu- roimaging, Springer, 2024, pp. 145–154

    2024

  16. [16]

    Jiang, N

    R. Jiang, N. Zuo, J. M. Ford, S. Qi, D. Zhi, C. Zhuo, Y. Xu, Z. Fu, J. Bustillo, J. A. Turner, V. D. Calhoun, J. Sui, Task-induced brain connectivity promotes the detection of individual differences in brain-behavior relationships, NeuroImage 207 (2020) 116370. doi:https://doi.org/10.1016/j.neuroimage.2019.116370

    work page doi:10.1016/j.neuroimage.2019.116370 2020

  17. [17]

    W. Zhao, C. Makowski, D. J. Hagler, H. P. Garavan, W. K. Thompson, D. J. Greene, T. L. Jernigan, A. M. 10 Dale, Task fmri paradigms may capture more behav- iorally relevant information than resting-state func- tional connectivity, NeuroImage 270 (2023) 119946. doi:https://doi.org/10.1016/j.neuroimage.2023.119946

    work page doi:10.1016/j.neuroimage.2023.119946 2023

  18. [18]

    T. N. Kipf, M. Welling, Semi-supervised classifi- cation with graph convolutional networks, CoRR abs/1609.02907 (2016). arXiv:1609.02907

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  19. [19]

    Veličković, G

    P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Liò, Y. Bengio, Graph attention networks, in: In- ternational Conference on Learning Representations, 2018

    2018

  20. [20]

    W. L. Hamilton, R. Ying, J. Leskovec, Induc- tive representation learning on large graphs, CoRR abs/1706.02216 (2017). arXiv:1706.02216

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  21. [21]

    Gadgil, Q

    S. Gadgil, Q. Zhao, A. Pfefferbaum, E. V. Sullivan, E. Adeli, K. M. Pohl, Spatio-temporal graph convo- lution for resting-state fmri analysis, Medical image computing and computer-assisted intervention (MIC- CAI) 12267 (2020) 528–538

    2020

  22. [22]

    X. Chen, H. Fan, R. B. Girshick, K. He, Improved baselines with momentum contrastive learning, CoRR abs/2003.04297 (2020). arXiv:2003.04297

    work page internal anchor Pith review Pith/arXiv arXiv 2003

  23. [23]

    T. Chen, S. Kornblith, M. Norouzi, G. E. Hinton, A simple framework for contrastive learning of vi- sual representations, CoRR abs/2002.05709 (2020). arXiv:2002.05709

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  24. [24]

    X. Chen, K. He, Exploring simple siamese rep- resentation learning, CoRR abs/2011.10566 (2020). arXiv:2011.10566

    work page arXiv 2011

  25. [25]

    https://doi.org/10.1016/j.neuroimage.2012.02.018

    D. Van Essen, K. Ugurbil, E. Auerbach, D. Barch, T. Behrens, R. Bucholz, A. Chang, L. Chen, M. Cor- betta, S. Curtiss, S. Della Penna, D. Feinberg, M. Glasser, N. Harel, A. Heath, L. Larson-Prior, D. Marcus, G. Michalareas, S. Moeller, R. Oosten- veld, S. Petersen, F. Prior, B. Schlaggar, S. Smith, A. Snyder, J. Xu, E. Yacoub, The human con- nectome proje...

    work page doi:10.1016/j.neuroimage.2012.02.018 2012

  26. [26]

    M. D. Kaiser, C. M. Hudac, S. Shultz, S. M. Lee, C. Cheung, A. M. Berken, B. Deen, N. B. Pitskel, D. R. Sugrue, A. C. Voos, C. A. Saulnier, P. Ventola, J. M. Wolf, A. Klin, B. C. V. Wyk, K. A. Pelphrey, Neural signatures of autism, Proceedings of the Na- tional Academy of Sciences 107 (49) (2010) 21223– 21228. doi:10.1073/pnas.1010412107

    work page doi:10.1073/pnas.1010412107 2010

  27. [27]

    D. Yang, K. A. Pelphrey, D. G. Sukhodolsky, M. J. Crowley, E. Dayan, N. C. Dvornek, A. Venkatara- man, J. Duncan, L. Staib, P. Ventola, et al., Brain responses to biological motion predict treatment out- come in young children with autism, Translational Psychiatry 6 (11) (2016). doi:10.1038/tp.2016.213

    work page doi:10.1038/tp.2016.213 2016

  28. [28]

    X. Shen, F. Tokoglu, X. Papademetris, R. Con- stable, Groupwise whole-brain parcellation from resting-state fmri data for network node identification, NeuroImage 82 (2013) 403–415. doi:https://doi.org/10.1016/j.neuroimage.2013.05.081

    work page doi:10.1016/j.neuroimage.2013.05.081 2013

  29. [29]

    R. S. Desikan, F. Ségonne, B. Fischl, B. T. Quinn, B. C. Dickerson, D. Blacker, R. L. Buck- ner, A. M. Dale, R. P. Maguire, B. T. Hyman, M. S. Albert, R. J. Killiany, An automated la- beling system for subdividing the human cerebral cortex on mri scans into gyral based regions of interest, NeuroImage 31 (3) (2006) 968–980. doi:https://doi.org/10.1016/j.ne...

    work page doi:10.1016/j.neuroimage.2006.01.021 2006

  30. [30]

    W. Ding, X. Shen, J. Huang, H. Ju, Y. Chen, T. Yin, Brain age prediction based on resting-state func- tional mri using similarity metric convolutional neu- ral network, IEEE Access 11 (2023) 57071–57082. doi:10.1109/ACCESS.2023.3283148

    work page doi:10.1109/access.2023.3283148 2023

  31. [31]

    X. Li, N. C. Dvornek, X. Papademetris, J. Zhuang, L. H. Staib, P. Ventola, J. S. Duncan, 2-channel convolutional 3d deep neural network (2cc3d) for fmri analysis: Asd classification and feature learn- ing, in: 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018), 2018, pp. 1252–

    2018

  32. [32]

    doi:10.1109/ISBI.2018.8363798

    work page doi:10.1109/isbi.2018.8363798 2018

  33. [33]

    T. Hahn, J. Ernsting, N. R. Winter, V. Holstein, R. Leenings, M. Beisemann, L. Fisch, K. Sarink, D. Emden, N. Opel, R. Redlich, J. Repple, D. Grote- gerd, S. Meinert, J. G. Hirsch, T. Niendorf, B. En- demann, F. Bamberg, T. Kröncke, R. Bülow, H. Völzke, O. von Stackelberg, R. F. Sowade, L. Umutlu, B. Schmidt, S. Caspers, H. Kugel, T. Kircher, B. Risse, C....

    work page arXiv 2021

  34. [34]

    H. Li, T. D. Satterthwaite, Y. Fan, Brain age prediction based on resting-state functional con- nectivity patterns using convolutional neural net- works, Proceedings. IEEE International Sympo- sium on Biomedical Imaging 2018 (2018) 101–104. doi:10.1109/ISBI.2018.8363532

    work page doi:10.1109/isbi.2018.8363532 2018

  35. [35]

    Y. Zhu, Y. Xu, F. Yu, Q. Liu, S. Wu, L. Wang, Deep graph contrastive representation learning (2020). arXiv:2006.04131

    work page arXiv 2020

  36. [36]

    R. Ying, D. Bourgeois, J. You, M. Zitnik, J. Leskovec, Gnnexplainer: Generating explanations for graph neural networks (2019). arXiv:1903.03894. 11

    work page arXiv 2019

  37. [37]

    X. Shen, E. S. Finn, D. Scheinost, M. D. Rosen- berg, M. M. Chun, X. Papademetris, R. T. Con- stable, Using connectome-based predictive model- ing to predict individual behavior from brain con- nectivity, Nature Protocols 12 (3) (2017) 506–518. doi:10.1038/nprot.2016.178

    work page doi:10.1038/nprot.2016.178 2017

  38. [38]

    C. Wang, Y. Jiang, Z. Peng, C. Li, C. Bang, L. Zhao, J. Lv, J. Sepulcre, C. Yang, L. He, T. Liu, D. Bar- ron, Q. Li, R. Hirschtick, B.-H. Kim, X. Li, Y. Yuan, Towards a general-purpose foundation model for fmri analysis (2025). arXiv:2506.11167

    work page arXiv 2025

  39. [39]

    Ortega Caro, A

    J. Ortega Caro, A. H. de Oliveira Fonseca, S. Rizvi, M. Rosati, C. Averill, J. Cross, P. Mittal, E. Zappala, R. Dhodapkar, C. Abdallah, D. van Dijk, BrainLM: A foundation model for brain activity recordings, in: B. Kim, Y. Yue, S. Chaudhuri, K. Fragkiadaki, M. Khan, Y. Sun (Eds.), International Conference on Representation Learning, Vol. 2024, 2024, pp. 565– 576

    2024

  40. [40]

    Z. Wei, T. Dan, G. Wu, Large connectome model: An fmri foundation model of brain connectomes em- powered by brain-environment interaction in multi- task learning landscape (2025). arXiv:2510.18910

    work page arXiv 2025

  41. [41]

    Tomasi, L

    D. Tomasi, L. Chang, E. Caparelli, T. Ernst, Sex differences in sensory gating of the thalamus during auditory interference of visual attention tasks, Neuroscience 151 (4) (2008) 1006–1015. doi:https://doi.org/10.1016/j.neuroscience.2007.08.040

    work page doi:10.1016/j.neuroscience.2007.08.040 2008

  42. [42]

    Kennepohl, V

    S. Kennepohl, V. Sziklas, K. Garver, D. Wag- ner, M. Jones-Gotman, Memory and the medial temporal lobe: Hemispheric specialization re- considered, NeuroImage 36 (3) (2007) 969–978. doi:https://doi.org/10.1016/j.neuroimage.2007.03.049

    work page doi:10.1016/j.neuroimage.2007.03.049 2007

  43. [43]

    S. Xu, M. Li, C. Yang, X. Fang, M. Ye, L. Wei, J. Liu, B. Li, Y. Gan, B. Yang, W. Huang, P. Li, X. Meng, Y. Wu, G. Jiang, Altered functional connectivity in childrenwithlow-functionautismspectrumdisorders, Frontiers in Neuroscience Volume 13 - 2019 (2019). doi:10.3389/fnins.2019.00806

    work page doi:10.3389/fnins.2019.00806 2019

  44. [44]

    H. Jeon, A. Hur, H. Lee, Y.-W. Shin, S.-I. Lee, C.- J. Shin, S. Kim, G. Ju, J. Lee, J. Jung, S. Chung, J.-W. Son, The relationship between brain acti- vation for taking others’ perspective and intero- ceptive abilities in autism spectrum disorder: An fmri study, Journal of the Korean Academy of Child and Adolescent Psychiatry 35 (2024) 197–209. doi:10.576...

    work page doi:10.5765/jkacap.240008 2024

  45. [45]

    Y. Xiao, A. Friederici, D. Margulies, J. Brauer, Longitudinal changes in resting-state fmri from age 5 to age 6 years covary with lan- guage development, NeuroImage 128 (12 2015). doi:10.1016/j.neuroimage.2015.12.008

    work page doi:10.1016/j.neuroimage.2015.12.008 2015

  46. [46]

    Attention Is All You Need

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkor- eit, L. Jones, A. N. Gomez, L. Kaiser, I. Polo- sukhin, Attention is all you need (2017). doi:10.48550/ARXIV.1706.03762

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1706.03762 2017

  47. [47]

    DINOv3

    O. Siméoni, H. V. Vo, M. Seitzer, F. Baldassarre, M. Oquab, C. Jose, V. Khalidov, M. Szafraniec, S. Yi, M. Ramamonjisoa, F. Massa, D. Haziza, L. Wehrst- edt, J. Wang, T. Darcet, T. Moutakanni, L. Sentana, C. Roberts, A. Vedaldi, J. Tolan, J. Brandt, C. Cou- prie, J. Mairal, H. Jégou, P. Labatut, P. Bojanowski, Dinov3 (2025). arXiv:2508.10104

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  48. [48]

    A. A. Chen, D. Srinivasan, R. Pomponio, Y. Fan, I. M. Nasrallah, S. M. Resnick, L. L. Beason- Held, C. Davatzikos, T. D. Satterthwaite, D. S. Bassett, R. T. Shinohara, H. Shou, Harmonizing functional connectivity reduces scanner effects in communitydetection, NeuroImage256(2022)119198. doi:https://doi.org/10.1016/j.neuroimage.2022.119198. 12

    work page doi:10.1016/j.neuroimage.2022.119198 2022