Pith. sign in

REVIEW 4 major objections 4 minor 44 references

DB-GNN: Dual-Branch Graph Neural Network with Multi-Level Contrastive Learning for Jointly Identifying Within- and Cross-Frequency Coupled Brain Networks

T0 review · 4 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read The paper reports that jointly modeling within- and cross-frequency brain couplings in one graph network reaches 97.88% accuracy on the SEED emotion dataset.

desk verdict A legitimate new architecture with a leak-suspicious SEED split; the 97.88% claim is unverifiable until the split is clarified. read the letter →

arxiv 2504.20744 v1 pith:TPWW4AXI submitted 2025-04-29 q-bio.NC

classification q-bio.NC
keywords EEGemotionrecognitionbrainnetworkscross-frequencycouplingwithin-frequencygraphneuralcontrastivelearningtransformerSEEDdataset
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Within-frequency coupling and cross-frequency coupling are usually analyzed separately in EEG emotion studies; this paper claims that reading both at once, with one network, works better than either alone. Its DB-GNN has a local branch that processes each coupling graph with graph attention and a global branch that processes all coupling graphs together with a Transformer whose attention is biased by known coupling strengths. Node- and graph-level contrastive losses tie the two branches together and act as regularization. On SEED the paper reports 97.88% test accuracy and 97.87% F1, which it calls state-of-the-art. A sympathetic reader would take away that joint modeling of the two coupling types, not just more parameters, is what drives the gain.

What carries the argument

The central mechanism is the prior information-based graph transformer module (PiGTM). It is a Transformer self-attention block whose logits are computed as the scaled query-key product masked by the adjacency matrix, plus a learned function of the PLV/MI coupling strength between nodes. That addition lets global attention be guided by known neurophysiological coupling rather than learned purely from labels. The second mechanism is multi-level contrastive regularization: the global and local branch representations of the same node are positive pairs, and the pooled graph representations of the same brain network are positive pairs, while other nodes and networks in the batch serve as negatives, with an InfoNCE loss. Together these force the local and global branches to agree on shared structure while remaining discriminative.

What would settle it

Run the methods in Table III under the exact subject-dependent 80/20 partition used for DB-GNN; if any of them reaches or exceeds 97.88% accuracy or 97.87% F1, the claim of state-of-the-art performance is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that within-frequency coupling (WFC) and cross-frequency coupling (CFC) are complementary views of the same emotional brain state, and a model that reads both together beats models that read either alone. DB-GNN operationalizes this by constructing five WFC graphs with phase locking values and ten CFC graphs with modulation indices, then passing the individual graphs through a graph attention network while a prior information-based graph transformer (PiGTM) reads the whole set. The two branches are tied by node-level and graph-level contrastive losses. On SEED, the paper reports 97.88% test accuracy and 97.87% F1 with 0.87 subject standard deviation, and asserts this is state-of-the-art. The ablations support the joint-reading claim: removing the global branch, removing the prior coupling injection, or removing the contrastive regularization each lowers accuracy.

Load-bearing premise

The state-of-the-art comparison in Table III assumes every earlier method was tested under the same subject-dependent 80/20 split of the same SEED data as DB-GNN, and the paper does not show this.

Editorial extensions

If this is right

  • Ablation Model 1 versus DB-GNN shows that dropping the global branch drops mean accuracy from 97.87% to 91.56%, so global coupling information carries much of the model's performance.
  • Ablation Model 2 versus Model 3 shows that injecting prior coupling strengths into attention raises accuracy from 93.17% to 95.75%, so the prior information is doing real work.
  • Adding graph contrastive regularization (Model 3 vs DB-GNN) raises accuracy from 95.75% to 97.87% and lowers subject standard deviation from 1.65 to 0.87.
  • On all 15 SEED subjects DB-GNN stays above 95% accuracy, while at least one baseline drops below 35% on some subjects, so the design appears more stable across individuals.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same prior-injection trick, adding a learned function of edge weights to Transformer attention logits, could transfer to any graph classification task with meaningful edge strengths, such as fMRI functional connectivity or protein interaction networks.
  • A natural testable extension is a subject-independent evaluation: the SEED results here are subject-dependent, so a leave-one-subject-out protocol would reveal whether the 0.87 standard deviation reflects robustness to new individuals or only to new trials from seen subjects.
  • The manually fixed 20% density threshold for binarizing PLV and MI is a tuning choice; an adaptive or learned threshold could sharpen the contrast between emotion categories and is a direct follow-up the paper does not explore.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes DB-GNN, a dual-branch graph neural network for EEG emotion recognition that jointly models within-frequency coupling (WFC) and cross-frequency coupling (CFC) brain networks. The global branch uses a Transformer-based module (PiGTM) that injects PLV/MI coupling strengths as priors into self-attention, while the local branch uses GAT on individual WFC/CFC graphs. A multi-level (node and graph) InfoNCE contrastive loss regularizes the two branches. On the SEED dataset, the method is reported to achieve 97.88% mean accuracy and 97.87% mean F1-score in a subject-dependent evaluation, and the paper claims state-of-the-art performance. The manuscript includes per-subject comparisons against six baselines, Wilcoxon significance tests, and an ablation study.

Significance. If the reported results are valid, the paper makes a useful contribution by demonstrating that jointly exploiting WFC and CFC graphs with a dual-branch architecture and contrastive regularization can improve EEG emotion recognition. The internal comparisons in Tables I and II follow one evaluation protocol and show consistent gains over the implemented baselines, and the ablation study in Table IV shows monotonic improvement as each proposed component is added. These are strengths. However, the central claims rest on two verification-dependent issues: the exact train/test split granularity and the protocol compatibility of the Table III comparison. Since no code is released and several load-bearing hyperparameters are omitted, the results cannot currently be independently reproduced. The contribution is therefore promising but not yet fully substantiated.

major comments (4)
  1. [Section IV.A and Table I] The evaluation protocol does not specify the granularity of the 80/20 subject-dependent split. The text states that non-overlapping 3-second windows are extracted and then that 80% of the data is used for training and 20% for testing. If this split is performed at the window level, temporally adjacent windows from the same SEED trial can be assigned to both training and test sets. EEG windows within a trial are strongly autocorrelated and contain stimulus-locked activity, so a classifier can exploit trial-level confounds rather than learning generalizable emotion-related patterns. This directly affects the headline 97.88% accuracy. The authors must state whether the split is window-level, trial-level, or session-level, and if it is window-level, the evaluation must be redone with trial-disjoint or session-disjoint splits before the SOTA claim can be accepted.
  2. [Table III and Section IV.A] The state-of-the-art comparison in Table III is not validated for protocol compatibility. No information is given about the train/test split, subject-dependence, window length, or evaluation protocol used by the cited methods. In particular, the closest competitor SAGN [15] reports 97.62±0.74, but the paper does not demonstrate that this number was obtained under the same subject-dependent 80/20 protocol used for DB-GNN. Without this information, the claim that DB-GNN 'reaches the state-of-the-art performance' is not supported. Add a protocol column to Table III or restrict the SOTA claim to methods with explicitly identical evaluation settings.
  3. [Section IV.A and Eq. (19)] Several load-bearing hyperparameters and preprocessing details are missing: the exact values of the graph density thresholds T1 and T2, the contrastive loss weight λ, the temperature T, the embedding dimensions, the number of attention heads/layers, and the grid-search ranges are not reported. It is also not stated whether T1 and T2 are selected using only the training portion of the data; if the 20% density threshold is computed on the full dataset before splitting, that is a further leakage path. Because no code is released, these omissions prevent independent verification of the numerical results and of the ablation conclusions.
  4. [Eq. (13) and Section III.B/C] The dimension of the prior coupling term q is inconsistent with the global-branch input. The text defines q ∈ R^{N×N} for a graph with N nodes, but the PiGTM global branch is applied to the block adjacency matrix G ∈ R^{(5N)×(5N)} containing five frequency bands and N channels. It is unclear how the pairwise PLV/MI priors are arranged for a 5N-node attention matrix: are the priors block-diagonal, are cross-frequency blocks included, and how does the N×N definition in Eq. (13) extend to the 5N×5N case? Eq. (12) also uses the same symbol K for both the generic adjacency matrix and the block matrix G. This ambiguity makes the core architecture difficult to reproduce.
minor comments (4)
  1. [Section III.A, Eqs. (2)-(4)] The displayed formulas for PLV and MI are corrupted in the submitted text; for example, Eqs. (3) and (4) contain uninterpretable character sequences. Please ensure the final PDF renders these definitions correctly, since they are the basis of all graph constructions.
  2. [Section IV.A and Table I] The list of baselines in the text names GCN, GAT, SuperGAT, AntiSymmetric, and pmlp, but Table I and Figure 4 also report DirGNN. Please clarify whether DirGNN is included in all experiments and how its configuration relates to the cited baselines.
  3. [Section III.C, Eq. (18)] The graph-level contrastive loss uses negative samples from the same batch, but the batch composition is not described. State whether a batch contains windows from multiple subjects and multiple trials, because this affects the validity of the negative samples and the interpretation of the contrastive loss.
  4. [Section IV.C and Fig. 6] The caption of Fig. 6 says 'confusion matrices of comparison models,' but the figure shows ablation variants; this should be corrected. In addition, the text says 'positive samples as neural samples,' which should be 'positive samples as neutral samples.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DB-GNN's emotion recognition result is benchmarked on an external dataset and its coupling prior is feature reuse, not a fitted quantity renamed as a prediction.

full rationale

The paper's derivation chain is self-contained and empirically grounded: raw EEG is band-pass filtered, PLV and MI are computed to build within-frequency and cross-frequency adjacency matrices, and these matrices feed a dual-branch GAT/Transformer architecture with contrastive regularization and a final classifier. The only place where input-derived values reappear inside the model is Eq. (13), where the same PLV/MI coupling strengths used to build the adjacency matrices are injected as an additive prior q into the self-attention computation. This is feature reuse or domain-prior conditioning, not circular derivation: the model outputs emotion labels, and no parameter fitted to a subset of data is subsequently renamed as a predicted quantity. The state-of-the-art claim is evaluated against SEED, an external benchmark dataset; the closest prior baseline SAGN [15] shares one author with the current paper but is an independently published TNNLS result with different co-authors and is used only as a comparison point, not as justification for the architecture or for the correctness of the method. Ambiguity about whether the 80/20 subject-dependent split is performed at window, trial, or session level is a legitimate experimental-validity concern, but it is a correctness and reproducibility risk, not a circularity of derivation. No self-definitional, fitted-input-as-prediction, self-citation-load-bearing, or ansatz-smuggling pattern is present in the manuscript.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper depends on standard assumptions about PLV/MI informativeness and thresholding, plus a subject-dependent evaluation protocol. No new physical or mathematical entities are introduced.

free parameters (4)
  • Graph density threshold for PLV/MI binarization (T1, T2) = 20% brain network density
    Equations (6)-(7) binarize PLV/MI values using thresholds manually set to keep 20% density, which determines the input graph topology.
  • Contrastive loss weight lambda
    Equation (19) combines classification loss and contrastive loss with a weight lambda; the value is described only as an artificially specified smaller parameter.
  • Temperature T in InfoNCE loss
    Equations (17)-(18) use a temperature T in the contrastive losses; the value is not specified.
  • Architecture hyperparameters (embedding dims, heads, layers, grid search ranges)
    Section IV.A says grid search is used and hyperparameters refer to prior works, but no ranges or final values are given.
assumptions (4)
  • domain assumption PLV and MI computed from EEG faithfully capture within- and cross-frequency neural coupling relevant to emotion.
    Section III.A constructs all brain networks from PLV/MI; if these measures are not informative, the entire input representation is invalid.
  • domain assumption Binarizing coupling matrices at 20% density preserves the discriminative topology.
    Equations (6)-(7) discard all coupling strength information except thresholded edges, which could remove discriminative detail.
  • domain assumption Subject-dependent 80/20 split is an appropriate basis for generalization claims.
    Section IV.A and IV.B claim strong generalization from low std across subjects, but the split is within-subject; no cross-subject evaluation is reported.
  • ad hoc to paper Prior coupling injection in Eq. (13) reduces overfitting without distorting attention.
    The paper asserts this benefit but provides no theoretical or empirical isolation beyond the Model 2 vs Model 3 ablation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DB-GNN: Dual-Branch Graph Neural Network with Multi-Level Contrastive Learning for Jointly Identifying Within- and Cross-Frequency Coupled Brain Networks." pith.science (2026). https://pith.science/paper/TPWW4AXI

@misc{pith2026250420744,
  author       = {Pith},
  title        = {Pith review of: DB-GNN: Dual-Branch Graph Neural Network with Multi-Level Contrastive Learning for Jointly Identifying Within- and Cross-Frequency Coupled Brain Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TPWW4AXI}},
  note         = {Machine review of arXiv:2504.20744}
}
read the original abstract

Within-frequency coupling (WFC) and cross-frequency coupling (CFC) in brain networks reflect neural synchronization within the same frequency band and cross-band oscillatory interactions, respectively. Their synergy provides a comprehensive understanding of neural mechanisms underlying cognitive states such as emotion. However, existing multi-channel EEG studies often analyze WFC or CFC separately, failing to fully leverage their complementary properties. This study proposes a dual-branch graph neural network (DB-GNN) to jointly identify within- and cross-frequency coupled brain networks. Firstly, DBGNN leverages its unique dual-branch learning architecture to efficiently mine global collaborative information and local cross-frequency and within-frequency coupling information. Secondly, to more fully perceive the global information of cross-frequency and within-frequency coupling, the global perception branch of DB-GNN adopts a Transformer architecture. To prevent overfitting of the Transformer architecture, this study integrates prior within- and cross-frequency coupling information into the Transformer inference process, thereby enhancing the generalization capability of DB-GNN. Finally, a multi-scale graph contrastive learning regularization term is introduced to constrain the global and local perception branches of DB-GNN at both graph-level and node-level, enhancing its joint perception ability and further improving its generalization performance. Experimental validation on the emotion recognition dataset shows that DB-GNN achieves a testing accuracy of 97.88% and an F1- score of 97.87%, reaching the state-of-the-art performance.

Figures

Figures reproduced from arXiv: 2504.20744 by the authors.

Figure 1
Figure 1. Within-frequency and cross-frequency coupling brain networks.  and denote the within-frequency coupling brain networks constructed from multi-channel EEG signals in the  and frequency bands, respectively. _ represents the cross-frequency coupling brain network between the  and frequency bands. Previous studies analyze , , or  brain networks separately. This study seeks to jointly extract the global collaborat… view at source ↗
Figure 2
Figure 2. Architecture of our proposed algorithm. (a) The structure of the self-attention mechanism in the prior information-based graph transformer module (PiGTM); (b) Dual-branch graph neural network (DB-GNN) [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. The contrastive learning in DB-GNN. (a) Node-level contrastive learning; (b) Graph-level contrastive learning [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 6
Figure 6. Figure 6: First, [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 4
Figure 4. Figure 4: Distribution of testing accuracy of SOTA baselines and DB-GNN in SEED dataset. Note: Sup., Ant., Dir., PML., and DS-. Represent SuperGAT, AntiSymmetric, DirGNN, pmlp and DB-GNN. TABLE I. TESTING ACCURACY OF DB-GNN AND SOTA BASELINES ON THE SEED DATASET. Subjects GCN GA…
Figure 5
Figure 5. Figure 5: Confusion matrices of SOTA baselines and DB-GNN on SEED dataset [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Confusion matrices of comparison models on SEED dataset. TABLE III. SOTA METHODS ON THE SEED DATASET. Studies Year Method TeAccØstd F1-scoreØstd Xu et al. [32] 2019 GIN 88.64Ø2.26 88.63Ø2.25 Zhong et al. [22] 2022 RGNN 94.24Ø5.95 --- Jiang et al. [33] 2023 EmoGT 95.02Ø…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 19 canonical work pages

  1. [15]

    Application of Graph Theory for Identifying Connectivity Patterns in Human Brain Networks: A Systematic Review,

    F. V. Farahani, W. Karwowski, and N. R. Lighth all, “Application of Graph Theory for Identifying Connectivity Patterns in Human Brain Networks: A Systematic Review,” Front. Neurosci., v ol. 13, Jun. 2019, doi: 10.3389/fnins.2019.00585

  2. [1]

    This approach fully leverages b oth WFC and CFC information of emotional brain states, thereby laying a foundation for accurate emotion classification

    Dual-branch learning architecture : DB-GNN employs a dual-branch learning architecture, utiliz ing Graph Attention Networks (GAT) [17] to separately extract local features from within-frequency and cross- frequency coupled brain networks, and leveraging a prior information-based graph transformer module (PiGTM) to capture collaborative global features acr...

  3. [2]

    Prior information-based graph transformer module (PiGTM) :To enhance the global feature perception capability of DB-GNN, this study proposes employing a Transformer architecture to enable collaborative global information perception across all WFC/CFC brain networks. To mitigate the risk of overfitting, prior coupling information of brain networks was inte...

  4. [3]

    # $ = 1 %&’()*+, -.,/0 − +,$-.,/02 / 1 % ∈ , , , , -20 where !

    Multi-level graph contrastive learning :The DB- GNN incorporated both graph-level and node-level graph contrastive learning to constrain the dual-branch learning architecture, thereby enhancing the abilit y of DB-GNN to collaboratively perceive global and local features of brain networks while mitigating the overfitting risk associated with considering on...

  5. [4]

    EEG Source Imaging : A Practical Review of the Analysis Steps,

    C. M. Michel and D. Brunet, “EEG Source Imaging : A Practical Review of the Analysis Steps,” Front. Neurol., vol. 10, Ap r. 2019, doi: 10.3389/fneur.2019.00325

  6. [5]

    Emotion Downregulation Targets Interoceptive Brain Regions While Emotion Upregulation Targets Other Af fective Brain Regions,

    J. Min et al., “Emotion Downregulation Targets Interoceptive Brain Regions While Emotion Upregulation Targets Other Af fective Brain Regions,” J. Neurosci., vol. 42, no. 14, pp. 2973–2 985, Apr. 2022, doi: 10.1523/JNEUROSCI.1865-21.2022

  7. [6]

    The functional role of cross-frequency coupling,

    R. T. Canolty and R. T. Knight, “The functional role of cross-frequency coupling,” Trends Cogn. Sci., vol. 14, no. 11, pp. 506–515, Nov. 2010, doi: 10.1016/j.tics.2010.09.001

  8. [7]

    J. M. Palva, S. Monto, S. Kulashekhar, and S. Palva, “Neuronal synchrony reveals working memory networks and predicts indivi dual memory TABLE IV. ABLATION STUDIES ON THE SEED DATASET . Subjects Model 1 Model 2 Model 3 DB-GNN 1 90.16 90.73 95.44 96.83 2 85.02 88.37 92.87 96.79 3 91.23 91.44 94.94 97.43 4 91.80 92.37 94.58 97.61 5 88.73 88.94 94.08 97.13 6...

Show all 44 references
  1. [8]

    Low-frequency neuronal oscillations as instruments of sensory selection,

    C. E. Schroeder and P. Lakatos, “Low-frequency neuronal oscillations as instruments of sensory selection,” Trends Neurosci., vol. 32, no. 1, pp. 9– 18, Jan. 2009, doi: 10.1016/j.tins.2008.09.012

  2. [9]

    Alpha-band oscillations, attentio n, and controlled access to stored information,

    W. Klimesch, “Alpha-band oscillations, attentio n, and controlled access to stored information,” Trends Cogn. Sci., vol. 16, no. 12, pp. 606–617, Dec. 2012, doi: 10.1016/j.tics.2012.10.007

  3. [10]

    Emotion recogni tion based on group phase locking value using convolutional neural network,

    G. Cui, X. Li, and H. Touyama, “Emotion recogni tion based on group phase locking value using convolutional neural network,” Sci. Rep., vol. 13, no. 1, p. 3769, Mar. 2023, doi: 10.1038/s41598-023-30458-6

  4. [11]

    The Theta-Gamma Ne ural Code,

    J. E. Lisman and O. Jensen, “The Theta-Gamma Ne ural Code,” Neuron, vol. 77, no. 6, pp. 1002–1016, Mar. 2013, doi: 10.1016/j.neuron.2013.03.007

  5. [12]

    EEG emotion recognition based on cross-frequency granger causality feature extraction and fusion in the left and right hemispheres,

    J. Zhang, X. Zhang, G. Chen, L. Huang, and Y. S un, “EEG emotion recognition based on cross-frequency granger causality feature extraction and fusion in the left and right hemispheres,” Fron t. Neurosci., vol. 16, Sep. 2022, doi: 10.3389/fnins.2022.974673

  6. [13]

    Graph Neural Network-Based EEG Classification : A Survey | IEEE Journals & Magazine | IEEE Xplore

    “Graph Neural Network-Based EEG Classification : A Survey | IEEE Journals & Magazine | IEEE Xplore.” Accessed: Dec. 01, 2024. [Online]. Available: https://ieeexplore.ieee.org/abstract/document/10403874

  7. [14]

    Colloquiu m: Multiscale modeling of brain network organization,

    C. Presigny and F. De Vico Fallani, “Colloquiu m: Multiscale modeling of brain network organization,” Rev. Mod. Phys., vo l. 94, no. 3, p. 031002, Aug. 2022, doi: 10.1103/RevModPhys.94.031002

  8. [16]

    EEG Emotion Classification Based on Graph Convolutional Network,

    Z. Fan, F. Chen, X. Xia, and Y. Liu, “EEG Emotion Classification Based on Graph Convolutional Network,” Appl. Sci., vol. 14, no. 2, Art. no. 2, Jan. 2024, doi: 10.3390/app14020726

  9. [17]

    EEG Emotion Recognition Based on Dynamic Graph Neural Networks,

    Y. Guo, C. Tang, H. Wu, and B. Chen, “EEG Emotion Recognition Based on Dynamic Graph Neural Networks,” in 2024 IEEE Int ernational Symposium on Circuits and Systems (ISCAS), May 2024 , pp. 1–5. doi: 10.1109/ISCAS58744.2024.10558424

  10. [18]

    SAGN: Sp arse Adaptive Gated Graph Neural Network With Graph Regularization for Identifying Dual- View Brain Networks,

    W. Xue, H. He, Y. Wang, and Y. Zhao, “SAGN: Sp arse Adaptive Gated Graph Neural Network With Graph Regularization for Identifying Dual- View Brain Networks,” IEEE Trans. Neural Netw. Learn. Syst., pp. 1–15, 2024, doi: 10.1109/TNNLS.2024.3438835

  11. [19]

    M. J. Hülsemann, E. Naumann, and B. Rasch, “Qu antification of Phase- Amplitude Coupling in Neuronal Oscillations: Compar ison of Phase- Locking Value, Mean Vector Length, Modulation Index , and Generalized-Linear-Modeling-Cross-Frequency-Coupling,” Front. Neurosci., vol. 13, ...

  12. [20]

    Graph Attention Networks,

    P. Veličković, G. Cucurull, A. Casanova, A. Ro mero, P. Liò, and Y. Bengio, “Graph Attention Networks,” Feb. 04, 2018, arXiv: arXiv:1710.10903. doi: 10.48550/arXiv.1710.10903

  13. [21]

    Multi-Granularity Analysis of Brain Networks Assembled With Intra-Frequency and Cross-Frequency Phase Coup ling for Human EEG After Stroke,

    B. Ren et al., “Multi-Granularity Analysis of Brain Networks Assembled With Intra-Frequency and Cross-Frequency Phase Coup ling for Human EEG After Stroke,” Front. Comput. Neurosci., vol. 1 6, Mar. 2022, doi: 10.3389/fncom.2022.785397

  14. [22]

    Review of Graph Neural Network in Text Classification,

    M. Malekzadeh, P. Hajibabaee, M. Heidari, S. Z ad, O. Uzuner, and J. H. Jones, “Review of Graph Neural Network in Text Classification,” in 2021 IEEE 12th Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON), Dec. 2021, pp. 0 084–0091. doi: 10.110...

  15. [23]

    Graph Neural Networks and Their Current Applications in Bioinformatics,

    X.-M. Zhang, L. Liang, L. Liu, and M.-J. Tang, “Graph Neural Networks and Their Current Applications in Bioinformatics,” Front. Genet., vol. 12, Jul. 2021, doi: 10.3389/fgene.2021.690049

  16. [24]

    Gra ph Neural Networks in Network Neuroscience,

    A. Bessadok, M. A. Mahjoub, and I. Rekik, “Gra ph Neural Networks in Network Neuroscience,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 45, no. 5, pp. 5833–5848, May 2023, doi: 10.1109/TPAMI.2022.3209686

  17. [25]

    EEG-Based Emo tion Recognition Using Regularized Graph Neural Networks,

    P. Zhong, D. Wang, and C. Miao, “EEG-Based Emo tion Recognition Using Regularized Graph Neural Networks,” IEEE Tran s. Affect. Comput., vol. 13, no. 3, pp. 1290–1301, Jul. 2022, doi: 10.1109/TAFFC.2020.2994159

  18. [26]

    Repre sentation Learning with Contrastive Predictive Coding,

    A. van den Oord, Y. Li, and O. Vinyals, “Repre sentation Learning with Contrastive Predictive Coding,” Jan. 22, 2019, arXiv: arXiv:1807.03748. doi: 10.48550/arXiv.1807.03748

  19. [27]

    A Graph Neur al Network for EEG- Based Emotion Recognition With Contrastive Learning and Generative Adversarial Neural Network Data Augmentation,

    S. S. Gilakjani and H. A. Osman, “A Graph Neur al Network for EEG- Based Emotion Recognition With Contrastive Learning and Generative Adversarial Neural Network Data Augmentation,” IEEE Access, vol. 12, pp. 113–130, 2024, doi: 10.1109/ACCESS.2023.3344476

  20. [28]

    Emotion recognition of EEG signals based on contrastive lea rning graph convolutional model,

    Y. Zhang, Y. Liao, W. Chen, X. Zhang, and L. H uang, “Emotion recognition of EEG signals based on contrastive lea rning graph convolutional model,” J. Neural Eng., vol. 21, no. 4, p. 046060, Aug. 2024, doi: 10.1088/1741-2552/ad7060

  21. [29]

    On the theory of filter ampli fiers,

    S. Butterworth, “On the theory of filter ampli fiers,” Wirel. Eng., vol. 7, no. 6, pp. 536–541, 1930

  22. [30]

    Investigating Criti cal Frequency Bands and Channels for EEG-Based Emotion Recognition with Dee p Neural Networks,

    W.-L. Zheng and B.-L. Lu, “Investigating Criti cal Frequency Bands and Channels for EEG-Based Emotion Recognition with Dee p Neural Networks,” IEEE Trans. Auton. Ment. Dev., vol. 7, n o. 3, pp. 162–175, Sep. 2015, doi: 10.1109/TAMD.2015.2431497

  23. [31]

    Semi-Supervised Cl assification with Graph Convolutional Networks,

    T. N. Kipf and M. Welling, “Semi-Supervised Cl assification with Graph Convolutional Networks,” Feb. 22, 2017, arXiv: arXiv:1609.02907. doi: 10.48550/arXiv.1609.02907

  24. [32]

    How to Find Your Friendly N eighborhood: Graph Attention Design with Self-Supervision,

    D. Kim and A. Oh, “How to Find Your Friendly N eighborhood: Graph Attention Design with Self-Supervision,” Apr. 11, 2 022, arXiv: arXiv:2204.04879. doi: 10.48550/arXiv.2204.04879

  25. [33]

    Edge Directionality Improves Lear ning on Heterophilic Graphs,

    E. Rossi, B. Charpentier, F. D. Giovanni, F. F rasca, S. Günnemann, and M. M. Bronstein, “Edge Directionality Improves Lear ning on Heterophilic Graphs,” in Proceedings of the Second Learning on Graphs Conference, PMLR, Apr. 2024, p. 25:1-25:27. Accessed: Jan. 17, 2025. [Onlin...

  26. [34]

    Graph Ne ural Networks are Inherently Good Generalizers: Insights by Bridging GNNs and MLPs,

    C. Yang, Q. Wu, J. Wang, and J. Yan, “Graph Ne ural Networks are Inherently Good Generalizers: Insights by Bridging GNNs and MLPs,” Aug. 04, 2023, arXiv: arXiv:2212.09034. doi: 10.48550/arXiv.2212.09034

  27. [35]

    Ho w Powerful are Graph Neural Networks?,

    K. Xu, W. Hu, J. Leskovec, and S. Jegelka, “Ho w Powerful are Graph Neural Networks?,” Feb. 22, 2019, arXiv: arXiv:1810 .00826. doi: 10.48550/arXiv.1810.00826

  28. [36]

    Elastic Graph Transformer Networks for EEG-Based Emotion Recognit ion,

    W.-B. Jiang, X. Yan, W.-L. Zheng, and B.-L. Lu , “Elastic Graph Transformer Networks for EEG-Based Emotion Recognit ion,” in ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Jun. 2023, p p. 1–5. doi: 10.1109/ICASSP49357.2023...

  29. [37]

    Variational Instance-Adaptive Graph for EEG Emotion Recognition,

    T. Song et al., “Variational Instance-Adaptive Graph for EEG Emotion Recognition,” IEEE Trans. Affect. Comput., vol. 14, no. 01, pp. 343–356, Jan. 2023, doi: 10.1109/TAFFC.2021.3064940

  30. [38]

    GMSS: Graph-Based Multi-Task Se lf-Supervised Learning for EEG Emotion Recognition,

    Y. Li et al., “GMSS: Graph-Based Multi-Task Se lf-Supervised Learning for EEG Emotion Recognition,” IEEE Trans. Affect. C omput., vol. 14, no. 3, pp. 2512–2525, Jul. 2023, doi: 10.1109/TAFFC.2022.3170428

  31. [39]

    MSFR-GCN: A Multi-Scale Feature Reconstruction Graph Convolutional Network for EEG Emotion and Cognition Recognition,

    D. Pan et al., “MSFR-GCN: A Multi-Scale Feature Reconstruction Graph Convolutional Network for EEG Emotion and Cognition Recognition,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 31, pp . 3245–3254, 2023, doi: 10.1109/TNSRE.2023.3304660

  32. [40]

    Emotio n Recognition Using Hierarchical Spatiotemporal Electroencephalogram In formation from Local to Global Brain Regions,

    D.-K. Jeong, H.-G. Kim, and J.-Y. Kim, “Emotio n Recognition Using Hierarchical Spatiotemporal Electroencephalogram In formation from Local to Global Brain Regions,” Bioengineering, vol. 10, no. 9, Art. no. 9, Sep. 2023, doi: 10.3390/bioengineering10091040

  33. [41]

    E EG emotion recognition using EEG-SWTNS neural network through EEG spectral image,

    M. Cai, J. Chen, C. Hua, G. Wen, and R. Fu, “E EG emotion recognition using EEG-SWTNS neural network through EEG spectral image,” Inf. Sci., vol. 680, p. 121198, Oct. 2024, doi: 10.1016/j.ins.2024.121198

  34. [42]

    Graph Convolutional Network Wi th Connectivity Uncertainty for EEG-Based Emotion Recognition,

    H. Gao et al., “Graph Convolutional Network Wi th Connectivity Uncertainty for EEG-Based Emotion Recognition,” IEE E J. Biomed. Health Inform., vol. 28, no. 10, pp. 5917–5928, Oct . 2024, doi: 10.1109/JBHI.2024.3416944

  35. [43]

    EEG-based emotion recognition using a temporal-difference minimizing neural network,

    X. Ju, M. Li, W. Tian, and D. Hu, “EEG-based emotion recognition using a temporal-difference minimizing neural network,” Cogn. Neurodyn., vol. 18, no. 2, pp. 405–416, Apr. 2024, doi: 10.1007/s11571-023-10004-w

  36. [44]

    An Efficient Graph Learning Sys tem for Emotion Recognition Inspired by the Cognitive Prior Graph o f EEG Brain Network,

    C. Li et al., “An Efficient Graph Learning Sys tem for Emotion Recognition Inspired by the Cognitive Prior Graph o f EEG Brain Network,” IEEE Trans. Neural Netw. Learn. Syst., pp . 1–15, 2024, doi: 10.1109/TNNLS.2024.3405663

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.