pith. sign in

arxiv: 2606.29200 · v1 · pith:3CYG6QDLnew · submitted 2026-06-28 · 💻 cs.LG

BrainRiem: Riemannian Prototype Learning for Source-Free Cross-Site Brain Network Diagnosis

Kunyu Zhang , Tianxiang Xu This is my paper

Pith reviewed 2026-06-30 08:58 UTC · model grok-4.3

classification 💻 cs.LG
keywords source-free domain adaptationRiemannian prototype learningSPD manifoldfunctional connectivityfMRI diagnosisLog-Euclidean metricDirichlet energybrain network analysis
0
0 comments X

The pith

BrainRiem learns Riemannian prototypes on the SPD manifold that serve as anchors for source-free brain network diagnosis across sites.

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

The paper introduces BrainRiem to handle domain shifts in multi-site fMRI studies without sharing source data, which privacy rules forbid. It learns compact prototypes via bi-level optimization that respects the geometry of symmetric positive definite matrices. The Log-Euclidean metric keeps prototypes valid while Dirichlet Energy calibration matches their spectral properties to real brain networks. These prototypes are sent alone to target sites and used to train local diagnostic models. Experiments on ABIDE and REST-meta-MDD show gains over prior source-free and domain-adaptation baselines, and the prototypes display connectivity patterns consistent with neuroscience observations.

Core claim

We propose BrainRiem, a source-free domain adaptation framework learning compact Riemannian brain prototypes via manifold-aware bi-level optimization. It employs the Log-Euclidean Metric to ensure prototypes remain valid SPD matrices, while Dirichlet Energy spectral calibration aligns their frequency characteristics with real brain networks. Only anonymized prototypes are transmitted to target sites, serving as stable anchors for training local models without source data access and reducing leakage under the evaluated attacks.

What carries the argument

Manifold-aware bi-level optimization that produces Riemannian prototypes on the SPD manifold via the Log-Euclidean Metric and Dirichlet Energy spectral calibration.

If this is right

  • Target sites can train diagnostic models using only the received prototypes as anchors without any source data access.
  • Privacy leakage is reduced because only anonymized prototypes leave the source site.
  • Performance exceeds state-of-the-art source-free, traditional, and graph domain adaptation methods on ABIDE and REST-meta-MDD across scanners and demographics.
  • The resulting prototypes exhibit connectivity patterns that align with established neuroscience findings.

Where Pith is reading between the lines

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

  • Geometric preservation of manifold structure appears necessary to avoid distorting diagnostic signals that Euclidean operations would corrupt.
  • The prototype-only transmission pattern could support larger multi-site collaborations without requiring data centralization.
  • If the calibration step generalizes, similar spectral alignment techniques might improve prototype stability in other manifold-valued medical data.

Load-bearing premise

The learned Riemannian prototypes retain enough diagnostic signal after Log-Euclidean optimization and Dirichlet Energy calibration to act as stable anchors for target-site training when no source data is present.

What would settle it

If target-site models trained using the transmitted prototypes show no accuracy gain over models trained without any anchors or with Euclidean prototypes on held-out ABIDE or REST-meta-MDD test sets, the central claim would be falsified.

Figures

Figures reproduced from arXiv: 2606.29200 by Kunyu Zhang, Tianxiang Xu.

Figure 1
Figure 1. Figure 1: Motivation for BrainRiem. (a) Domain shift from scanner heterogeneity de￾grades model performance. (b) Data sharing constraints in multi-site collaborations. (c) Euclidean operations distort SPD manifold geometry; Riemannian operations pre￾serve structure. (d) BrainRiem learns manifold-aware prototypes for source-free adap￾tation. as a promising avenue for objective diagnosis [45,49]. These methods utilize… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the BrainRiem framework. The method consists of two privacy￾compliant stages: (a) Source-site prototype learning via manifold-aware bi-level opti￾mization; (b) Target-site adaptation using transmitted prototypes as stable anchors. 3 Methodology The framework in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Statistical heterogeneity across multi-site datasets. ABIDE (a-c): (a) Sample size imbalance across sites. (b) Age distribution by scanner vendor (Siemens, Philips, GE). (c) Demographic diversity (IQ, gender ratio). REST-meta-MDD (d-f): (d) Class balance variations. (e) Extreme age shifts (S3: ∼20yrs vs. S25: ∼67yrs). (f) Gender distribution biases. 4 Experimental Setup and Results 4.1 Experimental Setting… view at source ↗
Figure 4
Figure 4. Figure 4: Ablation analysis (REST-meta-MDD, LOSO). (a-e) Prototype initialization comparison across five sites; Riemannian-aware initialization outperforms baselines. (f) Component importance: Log-Euclidean mapping removal causes largest performance drop. method StruRW (70.4%). On REST-meta-MDD, BrainRiem achieves 70.2% av￾erage accuracy across 23 sites, significantly outperforming NRC (63.7%) by 6.5% and StruRW (64… view at source ↗
Figure 5
Figure 5. Figure 5: Hyperparameter and spectral analysis (REST-meta-MDD, LOSO). (a-d) Sen￾sitivity of λ1-λ4 across five sites; framework shows robustness. (e) Optimal K = 4 prototypes. (f) Dirichlet Energy: learned prototypes align with source/target distribu￾tions, preventing high-frequency artifacts. 0 10 20 30 40 0 50 100 Rank (K) Accuracy (%) Raw FC Ours Random Chance Re-identification Rate (%) Rank (K) 0 50 100 0 50 100 … view at source ↗
Figure 6
Figure 6. Figure 6: Privacy auditing (NYU). (a) Re-identification: raw FC achieves 100% Rank-1, BrainRiem reduces to 3.8% (random-level). (b) Membership inference: raw AUC=0.99, BrainRiem AUC=0.53 (near-random). 5 Discussion 5.1 Biological Interpretability of Learned Prototypes To verify the biological validity of learned prototypes, we visualized their topo￾logical structures in [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Learned brain prototypes (ABIDE). (a) TD: DMN-dominated “Rich-Club” organization. (b) ASD: sensory-motor and limbic dominance. (c) Differential (ASD−TD): hyper-connectivity in sensory-motor, hypo-connectivity in DMN. core regions, which provides evidence for the “DMN-underconnectivity” hypoth￾esis (see Supplementary Material C.1–C.2 for complete analysis). 5.2 Spectral Calibration and Privacy Analysis As s… view at source ↗
read the original abstract

Multi-site functional MRI (fMRI) studies are essential for robust neuropsychiatric diagnosis yet suffer severe domain shifts from scanner heterogeneity, demographics, and site-specific acquisition protocols. Traditional domain adaptation requires concurrent source and target data access, violating clinical privacy regulations. Moreover, functional connectivity matrices lie on the Symmetric Positive Definite (SPD) manifold, where Euclidean operations cause geometric distortions corrupting diagnostic patterns. We propose BrainRiem, a source-free domain adaptation framework learning compact Riemannian brain prototypes via manifold-aware bi-level optimization. It employs the Log-Euclidean Metric to ensure prototypes remain valid SPD matrices, while Dirichlet Energy spectral calibration aligns their frequency characteristics with real brain networks. Only anonymized prototypes are transmitted to target sites, serving as stable anchors for training local models without source data access and reducing leakage under the evaluated attacks. Comprehensive experiments on ABIDE and REST-meta-MDD show BrainRiem consistently outperforms state-of-the-art source-free, traditional, and graph domain adaptation methods across diverse scanners and demographics. Notably, learned prototypes exhibit biologically interpretable connectivity patterns aligning with established neuroscience findings, validating the necessity of Riemannian geometry for brain network analysis.

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

3 major / 2 minor

Summary. The manuscript proposes BrainRiem, a source-free domain adaptation framework for multi-site fMRI brain network diagnosis. It learns compact Riemannian prototypes on the SPD manifold via bi-level optimization that combines Log-Euclidean projection with Dirichlet Energy spectral calibration; only the anonymized prototypes are transmitted to target sites to anchor local model training without any source data access. The central claims are consistent outperformance over source-free, traditional, and graph domain adaptation baselines on ABIDE and REST-meta-MDD across scanners and demographics, plus biological interpretability of the learned prototypes.

Significance. If the empirical claims are substantiated with proper controls, the work would address a practically important privacy constraint in clinical neuroimaging while respecting the SPD geometry of functional connectivity. The Riemannian formulation is well-motivated and the source-free transmission of prototypes is a clear operational advantage. No machine-checked proofs or parameter-free derivations are present, but the explicit handling of manifold geometry and the attempt at biological validation are positive features.

major comments (3)
  1. [Abstract, §4] Abstract and §4 (Experiments): the central claim of consistent outperformance is asserted without any quantitative tables, statistical tests, ablation results, or error analysis supplied in the manuscript text. This directly prevents evaluation of whether the prototypes function as stable anchors under domain shift.
  2. [§3.2] §3.2 (Bi-level optimization): the claim that Log-Euclidean projection plus Dirichlet Energy calibration preserves class-discriminative connectivity patterns is load-bearing for the source-free guarantee, yet no analysis, ablation on the energy weighting coefficient, or comparison against randomized/generic SPD matrices is provided to rule out collapse of diagnostic directions.
  3. [§3] §3 (Prototype transmission): the source-free protocol assumes the transmitted prototypes retain sufficient diagnostic signal after manifold projection; without a quantitative check that target-site performance degrades when prototypes are replaced by non-informative SPD matrices, the circularity concern (performance metrics reducing to quantities defined by the fitted prototypes) remains unaddressed.
minor comments (2)
  1. [§3.1] Notation for the Dirichlet Energy term and the prototype update rule should be introduced with explicit equations rather than prose descriptions.
  2. [§2] The manuscript should cite prior Riemannian SPD work on brain networks (e.g., Log-Euclidean applications in fMRI) to situate the geometric contribution.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback highlighting the need for stronger empirical substantiation of the claims. We address each major comment point-by-point below and will revise the manuscript to incorporate the requested analyses and controls.

read point-by-point responses

  1. Referee: [Abstract, §4] Abstract and §4 (Experiments): the central claim of consistent outperformance is asserted without any quantitative tables, statistical tests, ablation results, or error analysis supplied in the manuscript text. This directly prevents evaluation of whether the prototypes function as stable anchors under domain shift.

    Authors: While §4 presents the full experimental tables and comparisons on ABIDE and REST-meta-MDD, the abstract indeed summarizes without specific metrics. In the revision we will insert key quantitative results (e.g., mean accuracy gains with standard deviations) into the abstract and add statistical significance tests (paired t-tests or Wilcoxon signed-rank with p-values) plus error analysis to the experimental section. revision: yes

  2. Referee: [§3.2] §3.2 (Bi-level optimization): the claim that Log-Euclidean projection plus Dirichlet Energy calibration preserves class-discriminative connectivity patterns is load-bearing for the source-free guarantee, yet no analysis, ablation on the energy weighting coefficient, or comparison against randomized/generic SPD matrices is provided to rule out collapse of diagnostic directions.

    Authors: We agree that direct evidence is required. The revised manuscript will add an ablation on the Dirichlet Energy weighting coefficient together with comparisons against randomized and generic SPD matrices, demonstrating that the calibration step retains class-discriminative directions rather than inducing collapse. revision: yes

  3. Referee: [§3] §3 (Prototype transmission): the source-free protocol assumes the transmitted prototypes retain sufficient diagnostic signal after manifold projection; without a quantitative check that target-site performance degrades when prototypes are replaced by non-informative SPD matrices, the circularity concern (performance metrics reducing to quantities defined by the fitted prototypes) remains unaddressed.

    Authors: To directly address the circularity concern, the revision will include a controlled experiment in which the transmitted prototypes are replaced by non-informative SPD matrices (identity or random SPD) and report the resulting drop in target-site performance, confirming that diagnostic signal is carried by the learned prototypes. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained.

full rationale

The paper describes a source-free adaptation pipeline in which prototypes are learned on source data via Log-Euclidean projection and Dirichlet Energy calibration, then transmitted as fixed anchors for target-site training. No equation or claim reduces a reported performance metric to a quantity defined by the same fitted prototypes; the target evaluation uses independent data and the transmitted prototypes are treated as external inputs. No self-citation is invoked as a load-bearing uniqueness theorem, no ansatz is smuggled via prior work, and no known empirical pattern is merely renamed. The central claim therefore rests on the empirical behavior of the bi-level optimization rather than on definitional equivalence.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 1 invented entities

Central claim depends on the effectiveness of learned prototypes as domain anchors and on the preservation of diagnostic signal under the chosen Riemannian metric and spectral regularizer; both are fitted quantities whose independent grounding is not shown in the abstract.

free parameters (2)
  • Number and dimensionality of prototypes
    Hyperparameter controlling capacity of the compact representation transmitted to target sites.
  • Dirichlet Energy weighting coefficient
    Controls how strongly prototype frequency content is forced to match real brain networks.
axioms (2)
  • domain assumption Functional connectivity matrices are elements of the SPD manifold
    Invoked to justify why Euclidean operations distort diagnostic patterns.
  • standard math Log-Euclidean metric maps SPD matrices to a flat space while preserving positive-definiteness
    Used to guarantee that learned prototypes remain valid SPD matrices.
invented entities (1)
  • Riemannian brain prototypes no independent evidence
    purpose: Compact, anonymized reference objects that replace source data at target sites
    New learned objects introduced by the method; no external falsifiable prediction is stated.

pith-pipeline@v0.9.1-grok · 5726 in / 1281 out tokens · 44044 ms · 2026-06-30T08:58:15.868390+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

51 extracted references · 7 canonical work pages · 5 internal anchors

  1. [1]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition workshops

    Acharya, D., Huang, Z., Pani Paudel, D., Van Gool, L.: Covariance pooling for facial expression recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition workshops. pp. 367–374 (2018)

    2018

  2. [2]

    SIAM journal on matrix analysis and applications29(1), 328–347 (2007)

    Arsigny, V., Fillard, P., Pennec, X., Ayache, N.: Geometric means in a novel vector space structure on symmetric positive-definite matrices. SIAM journal on matrix analysis and applications29(1), 328–347 (2007)

    2007

  3. [3]

    Advances in Neural Information Processing Systems32(2019)

    Brooks, D., Schwander, O., Barbaresco, F., Schneider, J.Y., Cord, M.: Rieman- nian batch normalization for spd neural networks. Advances in Neural Information Processing Systems32(2019)

    2019

  4. [4]

    In: 2023 3rd International conference on electronic information engineering and computer communication (EIECC)

    Cheng, W., Long, Z., Ma, X., Li, C.: Spd manifold-based graph neural network for fault diagnosis of harmonic drive. In: 2023 3rd International conference on electronic information engineering and computer communication (EIECC). pp. 1–

    2023

  5. [5]

    Molecular psychiatry19(6), 659–667 (2014)

    Di Martino, A., Yan, C.G., Li, Q., Denio, E., Castellanos, F.X., Alaerts, K., An- derson, J.S., Assaf, M., Bookheimer, S.Y., Dapretto, M., et al.: The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain ar- chitecture in autism. Molecular psychiatry19(6), 659–667 (2014)

    2014

  6. [6]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Ding, N., Xu, Y., Tang, Y., Xu, C., Wang, Y., Tao, D.: Source-free domain adap- tation via distribution estimation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 7212–7222 (2022)

    2022

  7. [7]

    Advances in neural information processing systems 34, 2848–2860 (2021)

    Dong, J., Fang, Z., Liu, A., Sun, G., Liu, T.: Confident anchor-induced multi- source free domain adaptation. Advances in neural information processing systems 34, 2848–2860 (2021)

    2021

  8. [8]

    In: International conference on machine learning

    Finn, C., Abbeel, P., Levine, S.: Model-agnostic meta-learning for fast adaptation of deep networks. In: International conference on machine learning. pp. 1126–1135. PMLR (2017)

    2017

  9. [9]

    Journal of machine learning research17(59), 1–35 (2016)

    Ganin, Y., Ustinova, E., Ajakan, H., Germain, P., Larochelle, H., Laviolette, F., March, M., Lempitsky, V.: Domain-adversarial training of neural networks. Journal of machine learning research17(59), 1–35 (2016)

    2016

  10. [10]

    Nature Communications17, 1589 (2026) 16 K

    Gifford,A.T.,Cichy,R.M.,Naselaris,T.,Kay,K.:A7tfmridatasetofsyntheticim- ages for out-of-distribution modeling of vision. Nature Communications17, 1589 (2026) 16 K. Zhang and T. Xu

    2026

  11. [11]

    Advances in neural information processing systems30(2017)

    Hamilton, W., Ying, Z., Leskovec, J.: Inductive representation learning on large graphs. Advances in neural information processing systems30(2017)

    2017

  12. [12]

    In: Pro- ceedings of the AAAI conference on artificial intelligence

    Huang, Z., Van Gool, L.: A riemannian network for spd matrix learning. In: Pro- ceedings of the AAAI conference on artificial intelligence. vol. 31 (2017)

    2017

  13. [13]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Huang, Z., Wu, J., Van Gool, L.: Building deep networks on grassmann manifolds. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 32 (2018)

    2018

  14. [14]

    Brain Sciences12(5), 659 (2022)

    Jiang, Q., Zhang, Y., Zheng, K.: Motor imagery classification via kernel-based domain adaptation on an spd manifold. Brain Sciences12(5), 659 (2022)

    2022

  15. [15]

    IEEE Transactions on Neural Networks and Learning Systems (2023)

    Ju, C., Guan, C.: Graph neural networks on spd manifolds for motor imagery classification: A perspective from the time–frequency analysis. IEEE Transactions on Neural Networks and Learning Systems (2023)

    2023

  16. [16]

    NeuroImage146, 1038– 1049 (2017)

    Kawahara, J., Brown, C.J., Miller, S.P., Booth, B.G., Chau, V., Grunau, R.E., Zwicker, J.G., Hamarneh, G.: Brainnetcnn: Convolutional neural networks for brain networks; towards predicting neurodevelopment. NeuroImage146, 1038– 1049 (2017)

    2017

  17. [17]

    Semi-Supervised Classification with Graph Convolutional Networks

    Kipf, T.: Semi-supervised classification with graph convolutional networks. arXiv preprint arXiv:1609.02907 (2016)

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  18. [18]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Kundu, J.N., Kulkarni, A., Singh, A., Jampani, V., Babu, R.V.: Generalize then adapt: Source-free domain adaptive semantic segmentation. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 7046–7056 (2021)

    2021

  19. [19]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Li, G., Jampani, V., Sevilla-Lara, L., Sun, D., Kim, J., Kim, J.: Adaptive pro- totype learning and allocation for few-shot segmentation. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 8334–8343 (2021)

    2021

  20. [20]

    Medical Image Analysis74, 102233 (2021)

    Li, X., Zhou, Y., Dvornek, N., Zhang, M., Gao, S., Zhuang, J., Scheinost, D., Staib, L.H., Ventola, P., Duncan, J.S.: Braingnn: Interpretable brain graph neural network for fmri analysis. Medical Image Analysis74, 102233 (2021)

    2021

  21. [21]

    In: International conference on machine learning

    Liang, J., Hu, D., Feng, J.: Do we really need to access the source data? source hy- pothesis transfer for unsupervised domain adaptation. In: International conference on machine learning. pp. 6028–6039. PMLR (2020)

    2020

  22. [22]

    In: Euro- pean conference on computer vision

    Liu, J., Song, L., Qin, Y.: Prototype rectification for few-shot learning. In: Euro- pean conference on computer vision. pp. 741–756. Springer (2020)

    2020

  23. [23]

    In: International conference on machine learn- ing

    Liu, S., Li, T., Feng, Y., Tran, N., Zhao, H., Qiu, Q., Li, P.: Structural re-weighting improves graph domain adaptation. In: International conference on machine learn- ing. pp. 21778–21793. PMLR (2023)

    2023

  24. [24]

    Liu, Y., Sun, M., Zhang, Y., Wang, Y.: Information bottleneck based graph struc- turallearningforoodgeneralization.InformationProcessing&Management63(5), 104691 (2026)

    2026

  25. [25]

    On First-Order Meta-Learning Algorithms

    Nichol, A., Achiam, J., Schulman, J.: On first-order meta-learning algorithms. arXiv preprint arXiv:1803.02999 (2018)

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  26. [26]

    In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2017

    Parisot, S., Ktena, S.I., Ferrante, E., Lee, M., Moreno, R.G., Glocker, B., Rueck- ert, D.: Spectral graph convolutions for population-based disease prediction. In: Medical Image Computing and Computer Assisted Intervention – MICCAI 2017. pp. 177–185. Springer-Verlag, Berlin, Heidelberg (2017)

    2017

  27. [27]

    Rahman, A., Rahman, M.S., Mahdy, M.R.C.: 3c-gan: class-consistent cyclegan for malariadomainadaptationmodel.BiomedicalPhysics&EngineeringExpress7(5), 055002 (2021)

    2021

  28. [28]

    Ad- vances in neural information processing systems30(2017)

    Snell, J., Swersky, K., Zemel, R.: Prototypical networks for few-shot learning. Ad- vances in neural information processing systems30(2017)

    2017

  29. [29]

    In: European conference on computer vision

    Sun,B.,Saenko,K.:Deepcoral:Correlationalignmentfordeepdomainadaptation. In: European conference on computer vision. pp. 443–450. Springer (2016) BrainRiem: Source-Free Brain Network Diagnosis 17

    2016

  30. [30]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Sung, F., Yang, Y., Zhang, L., Xiang, T., Torr, P.H., Hospedales, T.M.: Learning to compare: Relation network for few-shot learning. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 1199–1208 (2018)

    2018

  31. [31]

    IEEE Transactions on Neural Networks and Learning Systems35(6), 7363–7375 (2024)

    Tang, H., Ma, G., Guo, L., Fu, X., Huang, H., Zhan, L.: Contrastive brain network learning via hierarchical signed graph pooling model. IEEE Transactions on Neural Networks and Learning Systems35(6), 7363–7375 (2024)

    2024

  32. [32]

    In: Proceedings of the AAAI conference on artificial intelligence

    Tang, H., Jia, K.: Discriminative adversarial domain adaptation. In: Proceedings of the AAAI conference on artificial intelligence. vol. 34, pp. 5940–5947 (2020)

    2020

  33. [33]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Tzeng, E., Hoffman, J., Saenko, K., Darrell, T.: Adversarial discriminative domain adaptation. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 7167–7176 (2017)

    2017

  34. [34]

    Briefings in Bioinfor- matics27(1), bbag038 (2026)

    Wang,Y.,Liang,V.,Yin,N.,Liu,S.,Segal,E.:Sgac:agraphneuralnetworkframe- work for imbalanced and structure-aware amp classification. Briefings in Bioinfor- matics27(1), bbag038 (2026)

    2026

  35. [35]

    DisRFM: Polar Riemannian Flow Matching for Structure-Preserving Graph Domain Adaptation

    Wang, Y., Liu, X., Wang, M., Gao, S., Yin, N.: Riemannian flow matching for disentangled graph domain adaptation. arXiv preprint arXiv:2602.00656 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  36. [36]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Wang, Y., Wang, M., Huang, Z., Liu, S., Yin, N.: Nested graph pseudo-label re- finement for noisy label domain adaptation learning. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 40, pp. 26697–26705 (2026)

    2026

  37. [37]

    arXiv preprint arXiv:2602.08431 (2026)

    Wang, Y., Zhang, K., Wang, M., Gao, S., Yin, N.: Usbd: Universal struc- tural basis distillation for source-free graph domain adaptation. arXiv preprint arXiv:2602.08431 (2026)

    work page arXiv 2026

  38. [38]

    When Brain Networks Travel: Learning Beyond Site

    Wang, Y., Zhang, K., Yang, Y., Wolfers, T., Wu, Y., Gao, S., Yin, N.: When brain networks travel: Learning beyond site. arXiv preprint arXiv:2605.06050 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  39. [39]

    In: Proceedings of the web conference 2020

    Wu, M., Pan, S., Zhou, C., Chang, X., Zhu, X.: Unsupervised domain adaptive graph convolutional networks. In: Proceedings of the web conference 2020. pp. 1457–1467 (2020)

    2020

  40. [40]

    Proceedings of the National Academy of Sciences116(18), 9078–9083 (2019)

    Yan, C.G., Chen, X., Li, L., Castellanos, F.X., Bai, T.J., Bo, Q.J., Cao, J., Chen, G.M., Chen, N.X., Chen, W., et al.: Reduced default mode network functional connectivity in patients with recurrent major depressive disorder. Proceedings of the National Academy of Sciences116(18), 9078–9083 (2019)

    2019

  41. [41]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Yang, S., Wang, Y., Van De Weijer, J., Herranz, L., Jui, S.: Generalized source-free domain adaptation. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 8978–8987 (2021)

    2021

  42. [42]

    Advances in neural in- formation processing systems34, 29393–29405 (2021)

    Yang, S., Van de Weijer, J., Herranz, L., Jui, S., et al.: Exploiting the intrinsic neighborhood structure for source-free domain adaptation. Advances in neural in- formation processing systems34, 29393–29405 (2021)

    2021

  43. [43]

    In: ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Process- ing (ICASSP)

    Yang,Y.,Guo,X.,Cai,G.,Ye,C.,Ma,T.:Topology-regularizedself-knowledgedis- tillation for transductive-inductive learning of brain disorder diagnosis. In: ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Process- ing (ICASSP). pp. 4045–4049. IEEE (2024)

    2024

  44. [44]

    IEEE Transactions on Medical Imaging43(1), 108–121 (2024)

    Yang, Y., Ye, C., Guo, X., Wu, T., Xiang, Y., Ma, T.: Mapping multi-modal brain connectome for brain disorder diagnosis via cross-modal mutual learning. IEEE Transactions on Medical Imaging43(1), 108–121 (2024)

    2024

  45. [45]

    IEEE Transactions on Medical Imaging43(11), 4004–4016 (2024)

    Yang, Y., Ye, C., Su, G., Zhang, Z., Chang, Z., Chen, H., Chan, P., Yu, Y., Ma, T.: Brainmass: Advancing brain network analysis for diagnosis with large-scale self- supervised learning. IEEE Transactions on Medical Imaging43(11), 4004–4016 (2024)

    2024

  46. [46]

    In: The eleventh international conference on learning representations (2023) 18 K

    You, Y., Chen, T., Wang, Z., Shen, Y.: Graph domain adaptation via theory- grounded spectral regularization. In: The eleventh international conference on learning representations (2023) 18 K. Zhang and T. Xu

    2023

  47. [47]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Zhang, H., Koniusz, P., Jian, S., Li, H., Torr, P.H.: Rethinking class relations: Absolute-relative supervised and unsupervised few-shot learning. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 9432–9441 (2021)

    2021

  48. [48]

    Modeling Higher-Order Brain Interactions via a Multi-View Information Bottleneck Framework for fMRI-based Psychiatric Diagnosis

    Zhang, K., Li, Q., Calhoun, V.D., Yu, S.: Modeling higher-order brain interactions via a multi-view information bottleneck framework for fmri-based psychiatric di- agnosis. arXiv preprint arXiv:2604.17713 (2026)

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  49. [49]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention

    Zhang, K., Li, Q., Yu, S.: Mvho-ib: Multi-view higher-order information bottle- neck for brain disorder diagnosis. In: International Conference on Medical Image Computing and Computer-Assisted Intervention. pp. 407–417. Springer (2025)

    2025

  50. [50]

    arXiv preprint arXiv:2602.21046 (2026)

    Zhang, K., Yang, Y., Zhang, J., Shi, X., Yu, S.: Pime: Prototype-based in- terpretable mcts-enhanced brain network analysis for disorder diagnosis. arXiv preprint arXiv:2602.21046 (2026)

    work page arXiv 2026

  51. [51]

    Medical Image Analysis90, 102932 (2023)

    Zhang, S., Chen, X., Shen, X., Ren, B., Yu, Z., Yang, H., Jiang, X., Shen, D., Zhou, Y., Zhang, X.Y.: A-gcl: Adversarial graph contrastive learning for fmri analysis to diagnose neurodevelopmental disorders. Medical Image Analysis90, 102932 (2023)

    2023