arxiv: 2605.06050 · v1 · submitted 2026-05-07 · 💻 cs.LG

Recognition: unknown

When Brain Networks Travel: Learning Beyond Site

Yingxu Wang , Kunyu Zhang , Yanwu Yang , Thomas Wolfers , Yujie Wu , Siyang Gao , Nan Yin

Authors on Pith no claims yet

Pith reviewed 2026-05-08 13:58 UTC · model grok-4.3

classification 💻 cs.LG

keywords brain networksfMRIcross-site generalizationout-of-distributionconfounder decouplingfunctional connectivitygraph neural networkstransient dynamics

0 comments

The pith

A framework for brain network analysis decouples site-specific biases to generalize diagnostic predictions across unseen scanning locations.

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

The paper aims to solve the problem of poor generalization in graph-based fMRI brain network models when applied to new hospitals or scanners. It proposes that site effects create misleading shortcuts in connectivity data, and that standard averaging hides important short-term brain activity changes. By separating these site biases, building a common scaffold of reliable connections, modeling dynamic pathways on that scaffold, and adaptively weighting for each person, the method achieves better performance on data from completely new sites. If successful, this would allow brain-based diagnostic tools to work reliably no matter where the scans are taken.

Core claim

CORE achieves cross-site robustness by first decoupling site-aware confounders to isolate a population scaffold of diagnostic edges, then using lightweight temporal descriptors to profile transient pathway dynamics organized as a line graph, and finally applying prior-guided subject-adaptive gating to balance population priors with individual variability, leading to consistent outperformance of baselines under leave-one-site-out protocols on ABIDE, REST-meta-MDD, SRPBS, and ABCD datasets with gains up to 6.7%.

What carries the argument

The CORE framework, which uses site-aware confounder decoupling to extract a cross-site population scaffold, lightweight temporal descriptors for transient dynamics on a line graph of edges, and prior-guided subject-adaptive gating to preserve individual variability.

If this is right

Brain disorder classification can be made more reliable when training and testing data come from different sites.
Performance gains hold across multiple real-world multi-site fMRI collections for conditions like autism and depression.
The method is robust to changes in how the brain is divided into regions for analysis.
Transient neurodynamics captured this way contribute to better modeling than static connectivity averages.

Where Pith is reading between the lines

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

Similar decoupling techniques might help in other medical imaging domains affected by scanner variations.
Identifying the scaffold edges could point to stable biomarkers less affected by acquisition differences.
Extending the temporal descriptors to more detailed models could further improve capture of brain state transitions.
The line graph approach for pathways may generalize to other graph-based domain adaptation tasks.

Load-bearing premise

That the effects of different scanning sites can be separated from the true disease-related brain connectivity patterns without distorting the diagnostic information.

What would settle it

Training on several sites and testing on a new one, but finding that performance does not improve over baselines or that removing the decoupling step does not decrease performance.

Figures

Figures reproduced from arXiv: 2605.06050 by Kunyu Zhang, Nan Yin, Siyang Gao, Thomas Wolfers, Yanwu Yang, Yingxu Wang, Yujie Wu.

**Figure 1.** Figure 1: (a) Site-conditioned variations. (b) Time-varying FC. Although existing methods achieve strong performance, they degrade substantially under real-world cross-site OOD settings [41, 42, 56, 53]. This degradation is primarily due to structured, site-conditioned variations in brain network data arising from scanner characteristics and demographic biases [59, 65, 51], as illustrated in view at source ↗

**Figure 2.** Figure 2: Overview of the proposed CORE. Site-Aware Confounder Decoupling mitigates site view at source ↗

**Figure 3.** Figure 3: t-SNE visualizations on ABIDE. We compare CORE with all baseline methods under the LOSO protocol on the ABIDE, REST-meta-MDD, SRPBS, and ABCD datasets in view at source ↗

**Figure 4.** Figure 4: Ablation results in (a, b), gate-budget sensitivity view at source ↗

**Figure 5.** Figure 5: Top 10 positive brain connections identified on ABIDE. To assess the neurobiological relevance of the learned connectivity patterns, view at source ↗

**Figure 6.** Figure 6: Ablation results in (a, b), gate-budget sensitivity view at source ↗

**Figure 7.** Figure 7: Top 10 positive scaffold edges on ABIDE under the OOD setting. Edge strength is measured view at source ↗

read the original abstract

Graph-based learning on functional magnetic resonance imaging (fMRI) has shown strong potential for brain network analysis. However, existing methods degrade under cross-site out-of-distribution (OOD) settings because site-conditioned confounders induce non-pathological shortcuts, while functional connectivity constructed by temporal averaging obscures transient neurodynamics, limiting generalization to unseen sites. In this paper, we propose Cross-site OOD Robust brain nEtwork (CORE), a unified framework for brain network learning across unseen sites. CORE first performs site-aware confounder decoupling to mitigate site-conditioned bias and extract a cross-site population scaffold of reproducible diagnostic connectivity edges. It then profiles transient pathway dynamics over this scaffold using lightweight temporal descriptors and organizes scaffold edges into a line graph for transferable pathway-level modeling. Finally, CORE introduces a prior-guided subject-adaptive gating mechanism that leverages scaffold-derived population priors while preserving subject-specific connectivity variability. Extensive experiments under leave-one-site-out evaluation on real-world datasets (ABIDE, REST-meta-MDD, SRPBS, and ABCD) show that CORE consistently outperforms state-of-the-art baselines, with up to 6.7% relative gain. Furthermore, CORE remains robust to atlas variations, maintaining performance gains across different brain parcellation schemes.

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.

Desk Editor's Note private letter to a colleague

CORE combines site confounder decoupling, line-graph transient modeling, and adaptive gating for cross-site fMRI, but the decoupling step lacks validation that it preserves diagnostic signals rather than removing them.

read the letter

The main takeaway is that this paper introduces CORE, a pipeline that first decouples site-specific biases to build a cross-site scaffold of connectivity edges, then profiles transient pathways on that scaffold using line graphs and lightweight descriptors, and finally applies prior-guided subject-adaptive gating. It tests the whole thing under leave-one-site-out on ABIDE, REST-meta-MDD, SRPBS, and ABCD and reports up to 6.7% relative gains over baselines. That addresses a genuine practical problem in deploying fMRI models when scanners and populations differ. The sequence of steps is a reasonable assembly of existing ideas from domain adaptation and graph learning, and the shift from averaged functional connectivity to transient descriptors is a useful direction. The experiments span several real datasets and check atlas robustness, which is better than many single-site studies. The soft spots sit mainly in the decoupling stage. The central assumption is that site confounders can be stripped away while keeping the pathological connectivity patterns intact. If disease effects covary with site demographics or acquisition details, that step could easily attenuate the signals needed for diagnosis, and the abstract gives no direct evidence like biomarker overlap or pre/post-decoupling comparisons to rule that out. There are also no ablations, statistical tests, or error analysis shown, so the performance numbers are difficult to interpret on their own. The math looks like a composition of standard operations rather than a new derivation. This is for people working on multi-site neuroimaging and OOD generalization in medical graphs. A reader who needs concrete ideas for handling scanner bias in brain networks would find the framework worth examining if the full methods and validation hold up. It deserves peer review because the application matters and the approach is internally coherent, even though it will need more evidence on signal retention and component contributions.

Referee Report

3 major / 2 minor

Summary. The paper proposes CORE, a unified framework for cross-site OOD robust brain network learning from fMRI. It performs site-aware confounder decoupling to extract a population scaffold of reproducible diagnostic connectivity edges, profiles transient pathway dynamics via lightweight temporal descriptors, organizes scaffold edges into a line graph for pathway-level modeling, and applies a prior-guided subject-adaptive gating mechanism. Under leave-one-site-out evaluation on ABIDE, REST-meta-MDD, SRPBS, and ABCD, CORE outperforms SOTA baselines with up to 6.7% relative gain and remains robust to atlas variations.

Significance. If the central claims hold, the work addresses an important practical barrier in clinical neuroimaging: poor generalization across acquisition sites due to scanner/protocol confounds and the loss of dynamic information in static functional connectivity. The multi-dataset leave-one-site-out protocol and atlas-robustness checks are positive elements that strengthen the generalization narrative. The framework's emphasis on preserving subject-specific variability while leveraging population priors could influence downstream applications in ASD and MDD diagnosis if the decoupling step demonstrably retains diagnostic signals.

major comments (3)

[Abstract / Method description] The core technical claim—that site-aware confounder decoupling cleanly separates non-pathological site bias from diagnostic connectivity edges while preserving pathology—lacks any reported validation (e.g., overlap with established biomarkers, statistical retention tests, or ablation on synthetic site-disease interactions). This assumption is load-bearing for the OOD generalization result.
[Experiments] No ablation studies, statistical significance tests, or error analysis are referenced to support the reported 6.7% relative gain or the contribution of each component (decoupling, temporal descriptors, line-graph modeling, gating). Without these, the data-to-claim link cannot be assessed.
[Abstract / Method description] The claim that lightweight temporal descriptors suffice to capture transient neurodynamics (replacing temporal averaging) is presented without quantitative comparison to standard dynamic FC methods or analysis of what temporal scales are actually recovered.

minor comments (2)

[Abstract] The abstract would benefit from a one-sentence statement of the precise mathematical form of the confounder decoupling operation.
[Method] Notation for the line-graph construction and the prior-guided gating should be introduced with explicit equations rather than descriptive prose only.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments and the positive assessment of the work's practical relevance. We address each major comment below and commit to revisions that directly strengthen the evidential basis for our claims.

read point-by-point responses

Referee: [Abstract / Method description] The core technical claim—that site-aware confounder decoupling cleanly separates non-pathological site bias from diagnostic connectivity edges while preserving pathology—lacks any reported validation (e.g., overlap with established biomarkers, statistical retention tests, or ablation on synthetic site-disease interactions). This assumption is load-bearing for the OOD generalization result.

Authors: We agree that explicit validation of the decoupling step is necessary to substantiate the central claim. The current manuscript relies on downstream OOD gains and cross-dataset consistency as indirect evidence. In the revision we will add: (i) quantitative overlap between the retained diagnostic edges and established biomarkers reported in the ASD/MDD literature, (ii) statistical retention tests (e.g., correlation of subject-level scaffold connectivity with clinical scores before/after decoupling), and (iii) controlled synthetic experiments that inject known site-disease interactions to measure separation fidelity. These additions will be placed in a new subsection of the Methods and Results. revision: yes
Referee: [Experiments] No ablation studies, statistical significance tests, or error analysis are referenced to support the reported 6.7% relative gain or the contribution of each component (decoupling, temporal descriptors, line-graph modeling, gating). Without these, the data-to-claim link cannot be assessed.

Authors: We acknowledge the absence of component-wise ablations, formal significance testing, and error analysis in the submitted version. The reported gains are currently supported only by aggregate leave-one-site-out accuracies. In the revised manuscript we will include: (i) systematic ablations that isolate the contribution of each module, (ii) paired statistical tests (Wilcoxon signed-rank or t-tests with correction) on the 6.7% relative improvement across the four datasets, and (iii) error analysis comprising per-site performance variance, confusion matrices, and failure-case inspection. These will be presented in an expanded Experiments section with corresponding tables and figures. revision: yes
Referee: [Abstract / Method description] The claim that lightweight temporal descriptors suffice to capture transient neurodynamics (replacing temporal averaging) is presented without quantitative comparison to standard dynamic FC methods or analysis of what temporal scales are actually recovered.

Authors: The manuscript positions the lightweight descriptors as a computationally efficient alternative to full dynamic functional connectivity, but does not provide head-to-head quantitative comparisons or scale-recovery analysis. We will add these elements in the revision: (i) direct performance comparison against sliding-window dynamic FC and other standard dynamic methods on a representative subset of each dataset, and (ii) analysis of the temporal scales recovered by the descriptors (e.g., autocorrelation structure and frequency content). The new results will be reported in the Experiments section alongside the existing leave-one-site-out tables. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical framework with data-driven steps

full rationale

The paper describes CORE as a sequence of algorithmic operations (site-aware confounder decoupling to extract a population scaffold, lightweight temporal descriptors for pathway dynamics, line-graph organization, and prior-guided adaptive gating) without any equations that define quantities in terms of their own fitted outputs or predictions. No self-citation chains are invoked to justify uniqueness theorems or ansatzes, and performance claims rest on leave-one-site-out empirical results across multiple datasets rather than reductions to inputs by construction. This matches the default expectation for non-circular empirical ML papers; the derivation chain remains 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 summary.

pith-pipeline@v0.9.0 · 5526 in / 1178 out tokens · 57326 ms · 2026-05-08T13:58:21.622253+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

69 extracted references · 9 canonical work pages · 4 internal anchors

[1]

Tracking whole-brain connectivity dynamics in the resting state.Cerebral cortex, 24(3):663–676, 2014

Elena A Allen, Eswar Damaraju, Sergey M Plis, Erik B Erhardt, Tom Eichele, and Vince D Calhoun. Tracking whole-brain connectivity dynamics in the resting state.Cerebral cortex, 24(3):663–676, 2014

2014
[2]

Invariant Risk Minimization

Martin Arjovsky, Léon Bottou, Ishaan Gulrajani, and David Lopez-Paz. Invariant risk mini- mization.arXiv preprint arXiv:1907.02893, 2019

work page internal anchor Pith review arXiv 1907
[3]

Graph neural networks in network neuroscience.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(5):5833– 5848, 2022

Alaa Bessadok, Mohamed Ali Mahjoub, and Islem Rekik. Graph neural networks in network neuroscience.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(5):5833– 5848, 2022

2022
[4]

Detecting shortcut learning for fair medical ai using shortcut testing.Nature communications, 14(1):4314, 2023

Alexander Brown, Nenad Tomasev, Jan Freyberg, Yuan Liu, Alan Karthikesalingam, and Jessica Schrouff. Detecting shortcut learning for fair medical ai using shortcut testing.Nature communications, 14(1):4314, 2023

2023
[5]

Out-of- distribution detection on graphs: A survey.arXiv preprint arXiv:2502.08105, 2025

Tingyi Cai, Yunliang Jiang, Yixin Liu, Ming Li, Changqin Huang, and Shirui Pan. Out-of- distribution detection on graphs: A survey.arXiv preprint arXiv:2502.08105, 2025

work page arXiv 2025
[6]

The adolescent brain cognitive development (abcd) study: imaging acquisition across 21 sites

Betty Jo Casey, Tariq Cannonier, May I Conley, Alexandra O Cohen, Deanna M Barch, Mary M Heitzeg, Mary E Soules, Theresa Teslovich, Danielle V Dellarco, Hugh Garavan, et al. The adolescent brain cognitive development (abcd) study: imaging acquisition across 21 sites. Developmental cognitive neuroscience, 32:43–54, 2018

2018
[7]

Neurocognitive characterization of behaviour and mental illness through time-varying brain network analysis.Nature Communications, 2026

Xiao Chang, Tianye Jia, Zening Fu, Shitong Xiang, Yunman Xia, Chao Xie, Jiyan Zou, Miao Cao, Jie Zhang, Tobias Banaschewski, et al. Neurocognitive characterization of behaviour and mental illness through time-varying brain network analysis.Nature Communications, 2026

2026
[8]

Mitigating site effects in covariance for machine learning in neuroimaging data.Human brain mapping, 43(4):1179–1195, 2022

Andrew A Chen, Joanne C Beer, Nicholas J Tustison, Philip A Cook, Russell T Shinohara, Haochang Shou, and Alzheimer’s Disease Neuroimaging Initiative. Mitigating site effects in covariance for machine learning in neuroimaging data.Human brain mapping, 43(4):1179–1195, 2022

2022
[9]

A whole brain fmri atlas generated via spatially constrained spectral clustering

R Cameron Craddock, G Andrew James, Paul E Holtzheimer III, Xiaoping P Hu, and Helen S Mayberg. A whole brain fmri atlas generated via spatially constrained spectral clustering. Human brain mapping, 33(8):1914–1928, 2012

1914
[10]

Interpretable graph neural networks for connectome-based brain disorder analysis

Hejie Cui, Wei Dai, Yanqiao Zhu, Xiaoxiao Li, Lifang He, and Carl Yang. Interpretable graph neural networks for connectome-based brain disorder analysis. InInternational conference on medical image computing and computer-assisted intervention, pages 375–385. Springer, 2022

2022
[11]

Aberrant striatal functional connectivity in children with autism.Biological psychiatry, 69(9):847–856, 2011

Adriana Di Martino, Clare Kelly, Rebecca Grzadzinski, Xi-Nian Zuo, Maarten Mennes, Maria Angeles Mairena, Catherine Lord, F Xavier Castellanos, and Michael P Milham. Aberrant striatal functional connectivity in children with autism.Biological psychiatry, 69(9):847–856, 2011

2011
[12]

The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism.Molecular psychiatry, 19(6):659–667, 2014

Adriana Di Martino, Chao-Gan Yan, Qingyang Li, Erin Denio, Francisco X Castellanos, Kaat Alaerts, Jeffrey S Anderson, Michal Assaf, Susan Y Bookheimer, Mirella Dapretto, et al. The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism.Molecular psychiatry, 19(6):659–667, 2014. 10

2014
[13]

Debiasing graph neural networks via learning disentangled causal substructure.Proceedings of the Conference on Neural Information Processing Systems, 35:24934–24946, 2022

Shaohua Fan, Xiao Wang, Yanhu Mo, Chuan Shi, and Jian Tang. Debiasing graph neural networks via learning disentangled causal substructure.Proceedings of the Conference on Neural Information Processing Systems, 35:24934–24946, 2022

2022
[14]

Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity.Nature neuroscience, 18(11):1664– 1671, 2015

Emily S Finn, Xilin Shen, Dustin Scheinost, Monica D Rosenberg, Jessica Huang, Marvin M Chun, Xenophon Papademetris, and R Todd Constable. Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity.Nature neuroscience, 18(11):1664– 1671, 2015

2015
[15]

Robust regression.An R and S-Plus companion to applied regression, 91(6), 2002

John Fox and Sanford Weisberg. Robust regression.An R and S-Plus companion to applied regression, 91(6), 2002

2002
[16]

Fc-hgnn: A heterogeneous graph neural network based on brain functional connectivity for mental disorder identification

Yuheng Gu, Shoubo Peng, Yaqin Li, Linlin Gao, and Yihong Dong. Fc-hgnn: A heterogeneous graph neural network based on brain functional connectivity for mental disorder identification. Information Fusion, 113:102619, 2025

2025
[17]

Network-centric perspectives on neurodegeneration: Mechanisms, biomarkers, and therapeutic pathways.Brain Network Disorders, 2026

Rohan Gupta. Network-centric perspectives on neurodegeneration: Mechanisms, biomarkers, and therapeutic pathways.Brain Network Disorders, 2026

2026
[18]

A spatiotemporal graph transformer approach for alzheimer’s disease diagnosis with rs-fmri.Computers in Biology and Medicine, 178:108762, 2024

Peng He, Zhan Shi, Yaping Cui, Ruyan Wang, Dapeng Wu, Alzheimer’s Disease Neuroimaging Initiative, et al. A spatiotemporal graph transformer approach for alzheimer’s disease diagnosis with rs-fmri.Computers in Biology and Medicine, 178:108762, 2024

2024
[19]

Robust regression using iteratively reweighted least-squares

Paul W Holland and Roy E Welsch. Robust regression using iteratively reweighted least-squares. Communications in Statistics-theory and Methods, 6(9):813–827, 1977

1977
[20]

Spatio-temporal learning and explaining for dynamic functional connectivity analysis: Application to depression.Journal of Affective Disorders, 364:266–273, 2024

Jinlong Hu, Jianmiao Luo, Ziyun Xu, Bin Liao, Shoubin Dong, Bo Peng, and Gangqiang Hou. Spatio-temporal learning and explaining for dynamic functional connectivity analysis: Application to depression.Journal of Affective Disorders, 364:266–273, 2024

2024
[21]

Resting-state functional connectivity in autism spectrum disorders: a review.Frontiers in psychiatry, 7:205, 2017

Jocelyn V Hull, Lisa B Dokovna, Zachary J Jacokes, Carinna M Torgerson, Andrei Irimia, and John Darrell Van Horn. Resting-state functional connectivity in autism spectrum disorders: a review.Frontiers in psychiatry, 7:205, 2017

2017
[22]

Sampling rate effects on resting state fmri metrics.Frontiers in neuroscience, 13:279, 2019

Niko Huotari, Lauri Raitamaa, Heta Helakari, Janne Kananen, Ville Raatikainen, Aleksi Rasila, Timo Tuovinen, Jussi Kantola, Viola Borchardt, Vesa J Kiviniemi, et al. Sampling rate effects on resting state fmri metrics.Frontiers in neuroscience, 13:279, 2019

2019
[23]

Impact of sampling rate on statistical signifi- cance for single subject fmri connectivity analysis.Human brain mapping, 40(11):3321–3337, 2019

Oliver James, Hyunjin Park, and Seong-Gi Kim. Impact of sampling rate on statistical signifi- cance for single subject fmri connectivity analysis.Human brain mapping, 40(11):3321–3337, 2019

2019
[24]

A survey of graph neural networks in real world: Imbalance, noise, privacy and ood challenges.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

Wei Ju, Siyu Yi, Yifan Wang, Zhiping Xiao, Zhengyang Mao, Hourun Li, Yiyang Gu, Yifang Qin, Nan Yin, Senzhang Wang, et al. A survey of graph neural networks in real world: Imbalance, noise, privacy and ood challenges.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

2025
[25]

Fbnetgen: Task-aware gnn-based fmri analysis via functional brain network generation

Xuan Kan, Hejie Cui, Joshua Lukemire, Ying Guo, and Carl Yang. Fbnetgen: Task-aware gnn-based fmri analysis via functional brain network generation. InInternational conference on medical imaging with deep learning, pages 618–637. PMLR, 2022

2022
[26]

Brain network trans- former.Proceedings of the Conference on Neural Information Processing Systems, 35:25586– 25599, 2022

Xuan Kan, Wei Dai, Hejie Cui, Zilong Zhang, Ying Guo, and Carl Yang. Brain network trans- former.Proceedings of the Conference on Neural Information Processing Systems, 35:25586– 25599, 2022

2022
[27]

Brainnetcnn: Convolutional neural networks for brain networks; towards predicting neurodevelopment.NeuroImage, 146:1038–1049, 2017

Jeremy Kawahara, Colin J Brown, Steven P Miller, Brian G Booth, Vann Chau, Ruth E Grunau, Jill G Zwicker, and Ghassan Hamarneh. Brainnetcnn: Convolutional neural networks for brain networks; towards predicting neurodevelopment.NeuroImage, 146:1038–1049, 2017

2017
[28]

Kipf and Max Welling

Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. InProceedings of the International Conference on Learning Representations, 2017. 11

2017
[29]

Spatio-temporal graph convolutional network for diagnosis and treatment response prediction of major depressive disorder from functional connectivity

Youyong Kong, Shuwen Gao, Yingying Yue, Zhenhua Hou, Huazhong Shu, Chunming Xie, Zhijun Zhang, and Yonggui Yuan. Spatio-temporal graph convolutional network for diagnosis and treatment response prediction of major depressive disorder from functional connectivity. Human brain mapping, 42(12):3922–3933, 2021

2021
[30]

Jung Min Lee, Sunghyun Kyeong, Eunjoo Kim, and Keun-Ah Cheon. Abnormalities of inter- and intra-hemispheric functional connectivity in autism spectrum disorders: a study using the autism brain imaging data exchange database.Frontiers in neuroscience, 10:191, 2016

2016
[31]

Out-of-distribution generalization on graphs: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

Haoyang Li, Xin Wang, Ziwei Zhang, and Wenwu Zhu. Out-of-distribution generalization on graphs: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

2025
[32]

Learning invariant graph representations for out-of-distribution generalization.Proceedings of the Conference on Neural Information Processing Systems, 35:11828–11841, 2022

Haoyang Li, Ziwei Zhang, Xin Wang, and Wenwu Zhu. Learning invariant graph representations for out-of-distribution generalization.Proceedings of the Conference on Neural Information Processing Systems, 35:11828–11841, 2022

2022
[33]

The linear l 1 estimator and the huber m-estimator.SIAM Journal on Optimization, 8(2):457–475, 1998

Wu Li and JJ Swetits. The linear l 1 estimator and the huber m-estimator.SIAM Journal on Optimization, 8(2):457–475, 1998

1998
[34]

Braingnn: Interpretable brain graph neural network for fmri analysis.Medical image analysis, 74:102233, 2021

Xiaoxiao Li, Yuan Zhou, Nicha Dvornek, Muhan Zhang, Siyuan Gao, Juntang Zhuang, Dustin Scheinost, Lawrence H Staib, Pamela Ventola, and James S Duncan. Braingnn: Interpretable brain graph neural network for fmri analysis.Medical image analysis, 74:102233, 2021

2021
[35]

Graph neural networks for brain graph learning: A survey.arXiv preprint arXiv:2406.02594, 2024

Xuexiong Luo, Jia Wu, Jian Yang, Shan Xue, Amin Beheshti, Quan Z Sheng, David McAlpine, Paul Sowman, Alexis Giral, and Philip S Yu. Graph neural networks for brain graph learning: A survey.arXiv preprint arXiv:2406.02594, 2024

work page arXiv 2024
[36]

Effect of scanning duration and sample size on reliability in resting state fmri dynamic causal modeling analysis.NeuroImage, 292:120604, 2024

Liangsuo Ma, Sarah E Braun, Joel L Steinberg, James M Bjork, Caitlin E Martin, Larry D Keen II, and F Gerard Moeller. Effect of scanning duration and sample size on reliability in resting state fmri dynamic causal modeling analysis.NeuroImage, 292:120604, 2024

2024
[37]

Interpretable and generalizable graph learning via stochastic attention mechanism

Siqi Miao, Mia Liu, and Pan Li. Interpretable and generalizable graph learning via stochastic attention mechanism. InProceedings of the International Conference on Machine Learning, pages 15524–15543. PMLR, 2022

2022
[38]

Individual variability in functional connectivity architecture of the human brain.Neuron, 77(3):586–595, 2013

Sophia Mueller, Danhong Wang, Michael D Fox, BT Thomas Yeo, Jorge Sepulcre, Mert R Sabuncu, Rebecca Shafee, Jie Lu, and Hesheng Liu. Individual variability in functional connectivity architecture of the human brain.Neuron, 77(3):586–595, 2013

2013
[39]

Adaptive graph learning with multi-graph convolutions for brain disorder classification

Fuad Noman, Raphaël C-W Phan, Hernando Ombao, and Chee-Ming Ting. Adaptive graph learning with multi-graph convolutions for brain disorder classification. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 56–65. Springer, 2025

2025
[40]

The dynamic functional connectome: State-of-the-art and perspectives.Neuroimage, 160:41–54, 2017

Maria Giulia Preti, Thomas AW Bolton, and Dimitri Van De Ville. The dynamic functional connectome: State-of-the-art and perspectives.Neuroimage, 160:41–54, 2017

2017
[41]

Metaexplainer: Revisit domain generalization of functional connectome analyses from the perspective of explainability.Medical Image Analysis, 105:103664, 2025

Xinmei Qiu, Yongheng Sun, Yilin Shi, Xujun Duan, Fan Wang, and Jianhua Ma. Metaexplainer: Revisit domain generalization of functional connectome analyses from the perspective of explainability.Medical Image Analysis, 105:103664, 2025

2025
[42]

Towards graph neural networks with domain-generalizable explainability for fmri-based brain disorder diagnosis

Xinmei Qiu, Fan Wang, Yongheng Sun, Chunfeng Lian, and Jianhua Ma. Towards graph neural networks with domain-generalizable explainability for fmri-based brain disorder diagnosis. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 454–464. Springer, 2024

2024
[43]

Deep coral: Correlation alignment for deep domain adaptation

Baochen Sun and Kate Saenko. Deep coral: Correlation alignment for deep domain adaptation. InProceedings of the European Conference on Computer Vision, pages 443–450. Springer, 2016

2016
[44]

Diverse and sparse mixture-of-experts for causal subgraph–based out-of-distribution graph learning

Jerry Sun, Mohamed Abubakr Hassan, Yaoyu Zhang, Wanying Zhang, and Chi-Guhn Lee. Diverse and sparse mixture-of-experts for causal subgraph–based out-of-distribution graph learning. InProceedings of the International Conference on Learning Representations, 2026. 12

2026
[45]

A multi-site, multi-disorder resting-state magnetic resonance image database.Scientific data, 8(1):227, 2021

Saori C Tanaka, Ayumu Yamashita, Noriaki Yahata, Takashi Itahashi, Giuseppe Lisi, Takashi Yamada, Naho Ichikawa, Masahiro Takamura, Yujiro Yoshihara, Akira Kunimatsu, et al. A multi-site, multi-disorder resting-state magnetic resonance image database.Scientific data, 8(1):227, 2021

2021
[46]

Brain networks and intelligence: a graph neural network based approach to resting state fmri data.Medical Image Analysis, 101:103433, 2025

Bishal Thapaliya, Esra Akbas, Jiayu Chen, Ram Sapkota, Bhaskar Ray, Pranav Suresh, Vince D Calhoun, and Jingyu Liu. Brain networks and intelligence: a graph neural network based approach to resting state fmri data.Medical Image Analysis, 101:103433, 2025

2025
[47]

Automated anatomical labeling of activations in spm using a macroscopic anatomical parcellation of the mni mri single-subject brain.Neuroimage, 15(1):273–289, 2002

Nathalie Tzourio-Mazoyer, Brigitte Landeau, Dimitri Papathanassiou, Fabrice Crivello, Octave Etard, Nicolas Delcroix, Bernard Mazoyer, and Marc Joliot. Automated anatomical labeling of activations in spm using a macroscopic anatomical parcellation of the mni mri single-subject brain.Neuroimage, 15(1):273–289, 2002

2002
[48]

Graph attention networks

Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks. InProceedings of the International Conference on Learning Representations, 2018

2018
[49]

Environment promoted invariant information learning for graph out-of-distribution generalization.Artificial Intelligence, page 104522, 2026

Shuo Wang, Mingchen Sun, Qiang Huang, and Ying Wang. Environment promoted invariant information learning for graph out-of-distribution generalization.Artificial Intelligence, page 104522, 2026

2026
[50]

Sgac: a graph neural network framework for imbalanced and structure-aware amp classification.Briefings in Bioinformatics, 27(1):bbag038, 2026

Yingxu Wang, Victor Liang, Nan Yin, Siwei Liu, and Eran Segal. Sgac: a graph neural network framework for imbalanced and structure-aware amp classification.Briefings in Bioinformatics, 27(1):bbag038, 2026

2026
[51]

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

Yingxu Wang, Xinwang Liu, Mengzhu Wang, Siyang Gao, and Nan Yin. Riemannian flow matching for disentangled graph domain adaptation.arXiv preprint arXiv:2602.00656, 2026

work page internal anchor Pith review Pith/arXiv arXiv 2026
[52]

Nested graph pseudo- label refinement for noisy label domain adaptation learning

Yingxu Wang, Mengzhu Wang, Zhichao Huang, Suyu Liu, and Nan Yin. Nested graph pseudo- label refinement for noisy label domain adaptation learning. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 26697–26705, 2026

2026
[53]

Degree-conscious spiking graph for cross-domain adaptation.arXiv preprint arXiv:2410.06883, 2024

Yingxu Wang, Mengzhu Wang, Houcheng Su, Nan Yin, Quanming Yao, and James Kwok. Degree-conscious spiking graph for cross-domain adaptation.arXiv preprint arXiv:2410.06883, 2024

work page arXiv 2024
[54]

DSBD: Dual-Aligned Structural Basis Distillation for Graph Domain Adaptation

Yingxu Wang, Kunyu Zhang, Jiaxin Huang, Mengzhu Wang, Mingyan Xiao, Siyang Gao, and Nan Yin. Dsbd: Dual-aligned structural basis distillation for graph domain adaptation.arXiv preprint arXiv:2604.03154, 2026

work page internal anchor Pith review Pith/arXiv arXiv 2026
[55]

Protomol: enhancing molecular property prediction via prototype-guided multimodal learning.Briefings in Bioinformatics, 26(6):bbaf629, 2025

Yingxu Wang, Kunyu Zhang, Jiaxin Huang, Nan Yin, Siwei Liu, and Eran Segal. Protomol: enhancing molecular property prediction via prototype-guided multimodal learning.Briefings in Bioinformatics, 26(6):bbaf629, 2025

2025
[56]

Usbd: Universal structural basis distillation for source-free graph domain adaptation

Yingxu Wang, Kunyu Zhang, Mengzhu Wang, Siyang Gao, and Nan Yin. Usbd: Universal struc- tural basis distillation for source-free graph domain adaptation.arXiv preprint arXiv:2602.08431, 2026

work page arXiv 2026
[57]

BrainOOD: Out-of-distribution generalizable brain network analysis

Jiaxing Xu, Yongqiang Chen, Xia Dong, Mengcheng Lan, Tiancheng HUANG, Qingtian Bian, James Cheng, and Yiping Ke. BrainOOD: Out-of-distribution generalizable brain network analysis. InProceedings of the International Conference on Learning Representations, 2025

2025
[58]

How powerful are graph neural networks? InProceedings of the International Conference on Learning Representations, 2019

Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural networks? InProceedings of the International Conference on Learning Representations, 2019

2019
[59]

Compu- tational mechanisms of neuroimaging biomarkers uncovered by multicenter resting-state fmri connectivity variation profile.Molecular Psychiatry, 30(11):5463–5474, 2025

Okito Yamashita, Ayumu Yamashita, Yuji Takahara, Yuki Sakai, Yasumasa Okamoto, Go Okada, Masahiro Takamura, Motoaki Nakamura, Takashi Itahashi, Takashi Hanakawa, et al. Compu- tational mechanisms of neuroimaging biomarkers uncovered by multicenter resting-state fmri connectivity variation profile.Molecular Psychiatry, 30(11):5463–5474, 2025

2025
[60]

Reduced default mode network functional connectivity in patients with recurrent major depressive disorder.Proceedings of the National Academy of Sciences, 116(18):9078–9083, 2019

Chao-Gan Yan, Xiao Chen, Le Li, Francisco Xavier Castellanos, Tong-Jian Bai, Qi-Jing Bo, Jun Cao, Guan-Mao Chen, Ning-Xuan Chen, Wei Chen, et al. Reduced default mode network functional connectivity in patients with recurrent major depressive disorder.Proceedings of the National Academy of Sciences, 116(18):9078–9083, 2019. 13

2019
[61]

Empowering graph invariance learning with deep spurious infomax

Tianjun Yao, Yongqiang Chen, Zhenhao Chen, Kai Hu, Zhiqiang Shen, and Kun Zhang. Empowering graph invariance learning with deep spurious infomax. InProceedings of the International Conference on Machine Learning, 2024

2024
[62]

Learning graph invariance by harnessing spuriosity

Tianjun Yao, Yongqiang Chen, Kai Hu, Tongliang Liu, Kun Zhang, and Zhiqiang Shen. Learning graph invariance by harnessing spuriosity. InProceedings of the International Conference on Learning Representations, 2025

2025
[63]

Xing, and Zhiqiang Shen

Tianjun Yao, Haoxuan Li, Yongqiang Chen, Tongliang Liu, Le Song, Eric P. Xing, and Zhiqiang Shen. Pruning spurious subgraphs for graph out-of-distribution generalization. InProceedings of the Conference on Neural Information Processing Systems, 2025

2025
[64]

Moving beyond functional connectivity: Time-series modeling for fmri-based brain disorder classification.IEEE Transactions on Medical Imaging, 2026

Guoqi Yu, Xiaowei Hu, Angelica I Aviles-Rivero, Anqi Qiu, and Shujun Wang. Moving beyond functional connectivity: Time-series modeling for fmri-based brain disorder classification.IEEE Transactions on Medical Imaging, 2026

2026
[65]

Statistical harmonization corrects site effects in functional connectivity measurements from multi-site fmri data.Human brain mapping, 39(11):4213–4227, 2018

Meichen Yu, Kristin A Linn, Philip A Cook, Mary L Phillips, Melvin McInnis, Maurizio Fava, Madhukar H Trivedi, Myrna M Weissman, Russell T Shinohara, and Yvette I Sheline. Statistical harmonization corrects site effects in functional connectivity measurements from multi-site fmri data.Human brain mapping, 39(11):4213–4227, 2018

2018
[66]

Kexin Zhang, Shuhan Liu, Song Wang, Weili Shi, Chen Chen, Pan Li, Sheng Li, Jundong Li, and Kaize Ding. A survey of deep graph learning under distribution shifts: from graph out-of-distribution generalization to adaptation.ACM Transactions on Knowledge Discovery from Data, 20(2):1–38, 2026

2026
[67]

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

Kunyu Zhang, Qiang Li, Vince D Calhoun, and Shujian Yu. Modeling higher-order brain interactions via a multi-view information bottleneck framework for fmri-based psychiatric diagnosis.arXiv preprint arXiv:2604.17713, 2026

work page internal anchor Pith review Pith/arXiv arXiv 2026
[68]

Mvho-ib: Multi-view higher-order information bottle- neck for brain disorder diagnosis

Kunyu Zhang, Qiang Li, and Shujian Yu. Mvho-ib: Multi-view higher-order information bottle- neck for brain disorder diagnosis. InInternational Conference on Medical Image Computing and Computer-Assisted Intervention, pages 407–417. Springer, 2025

2025
[69]

Pime: Prototype-based interpretable mcts-enhanced brain network analysis for disorder diagnosis.arXiv preprint arXiv:2602.21046, 2026

Kunyu Zhang, Yanwu Yang, Jing Zhang, Xiangjie Shi, and Shujian Yu. Pime: Prototype-based interpretable mcts-enhanced brain network analysis for disorder diagnosis.arXiv preprint arXiv:2602.21046, 2026. 14 A Algorithm Algorithm 1Framework of Cross-site OOD Robust brain nEtwork (CORE) Require: Training environments Etr with datasets {bDe}e∈Etr, unseen test ...

work page arXiv 2026