Pith. sign in

REVIEW 4 major objections 6 minor 5 cited by

AdaBrain-Bench: Benchmarking Brain Foundation Models for Brain-Computer Interface Applications

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Brain foundation models beat traditional EEG decoders in a new 13-dataset benchmark.

desk verdict Useful EEG foundation-model benchmark with a real data-leakage caveat: pretraining sets overlap several evaluation datasets, so the headline cross-subject gains need re-quantification before firm conclusions. read the letter →

arxiv 2507.09882 v2 pith:57SEPODV submitted 2025-07-14 cs.LG

Jiamin Wu , Zichen Ren , Junyu Wang , Pengyu Zhu , Yonghao Song , Mianxin Liu , Qihao Zheng , Lei Bai
show 2 more authors
Wanli Ouyang Chunfeng Song
This is my paper · ORCID
classification cs.LG
keywords brainfoundationmodelsEEGbrain-computerinterfacebenchmarkcross-subjectgeneralizationfew-shottransferself-supervisedpretrainingscore
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

The paper argues that self-supervised 'brain foundation models'—large neural networks pretrained on unlabeled EEG recordings—can improve how well decoders generalize to people they have never seen, and it builds a standardized benchmark to test that claim across many tasks. The benchmark, AdaBrain-Bench, spans 13 EEG datasets covering seven brain-computer interface applications, from emotion recognition and motor imagery to sleep staging and seizure detection. Comparing four public foundation models against four trained-from-scratch deep learning baselines, the paper reports that the two best foundation models beat the best traditional model in the cross-subject setting (64.61% and 62.66% vs 58.12% macro-average), with the largest gains when training data is scarce. The central practical message is that large-scale pretraining on heterogeneous EEG, plus flexible handling of different channel layouts, carries the transfer advantage.

What carries the argument

The load-bearing object is AdaBrain-Bench itself: a modular adaptation pipeline that couples standardized EEG preprocessing, plug-in task heads for classification, regression, and retrieval, and a transfer score defined as $TS = \lambda \cdot (P_{\mathrm{prt}} - P_{\mathrm{scr}})/P_{\mathrm{scr}} + (1-\lambda)\cdot (P_{\mathrm{prt}} - P_{\mathrm{scr}})/(P_{\mathrm{oracle}} - P_{\mathrm{scr}})$. The transfer score measures a pretrained model's gain over its own trained-from-scratch twin, isolating the value of pretraining from architecture. The benchmark's three evaluation settings—cross-subject, multi-subject, and few-shot—are designed to separate the challenges of inter-subject variability, intra-subject session drift, and data scarcity.

What would settle it

Re-run the cross-subject evaluation after removing from the benchmark every dataset that appears in a model's pretraining list (SEED, SEED-IV, TUEV, TUAB, Siena, SHHS, and related TUH sets) and check whether LaBraM and CBraMod still beat the best trained-from-scratch baseline by more than six points on the remaining tasks.

Watch

Extended reading notes

Core claim

The central discovery is that pretrained EEG foundation models—specifically LaBraM and CBraMod, both trained with masked-signal self-supervision on thousands of hours of EEG—show measurably better cross-subject generalization than supervised deep learning baselines trained from scratch, and that the advantage grows as the downstream training set shrinks. The paper establishes this through AdaBrain-Bench, which standardizes preprocessing, adaptation via full fine-tuning or linear probing, and evaluation across cross-subject, multi-subject, and few-shot settings. A transfer score, defined as the relative gain over a same-architecture model trained from scratch, is introduced to quantify the pretraining benefit; CBraMod records the highest few-shot transfer scores. The authors attribute the success of the top models to their larger and more diverse pretraining corpora and to channel-flexible embeddings that adapt to heterogeneous EEG montages.

Load-bearing premise

The benchmark treats held-out subjects on datasets that were part of each model's pretraining corpus as 'unseen,' so the reported cross-subject advantage could partly reflect familiarity with the data distributions rather than true generalization to new people.

Editorial extensions

If this is right

  • If the result holds, fine-tuning a channel-flexible foundation model should become the default starting point for new EEG decoding tasks, especially when labeled data are limited.
  • The transfer score gives researchers a way to compare pretraining strategies independently of architecture, so future models can be optimized for transferability rather than raw accuracy alone.
  • The finding that cross-subject performance lags multi-subject performance across all models suggests that inter-subject variability, not session drift, is the main barrier to real-world BCI deployment.
  • The observation that performance plateaus around 40 training subjects on sleep staging datasets gives a concrete data-collection target for cross-subject studies.
  • Full fine-tuning generally beats linear probing for current foundation models, implying that frozen representations are not yet linearly separable and that further pretraining improvements are needed for efficient adaptation.

Reading between the lines

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

  • The benchmark should be re-run with a leave-pretraining-corpus-out protocol, since several evaluated datasets overlap with the models' pretraining corpora; this would separate genuine cross-subject generalization from domain familiarity.
  • A natural extension is stress-testing foundation models on tasks with channel counts, sampling rates, and recording devices never seen in pretraining, because the paper's own analysis identifies channel compatibility as a key success factor.
  • The transfer score could be made more decision-relevant by reporting it alongside per-dataset confidence intervals, so downstream users know when a positive transfer score is within noise.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces AdaBrain-Bench, a standardized benchmark for non-invasive EEG-based BCI decoding that spans 13 datasets across 7 tasks. The benchmark evaluates four publicly available brain foundation models (BIOT, EEGPT, LaBraM, CBraMod) against four traditional supervised models (EEGNet, LDMA, ST-Tran, Conformer) under three transfer settings: cross-subject, multi-subject, and few-shot. The authors report that LaBraM and CBraMod outperform traditional baselines on macro-averaged metrics in the cross-subject setting, and that pretrained models show larger gains over their from-scratch counterparts when training data are scarce. They also introduce a transfer score to quantify the benefit of pretraining, and analyze the effects of adaptation strategy, normalization, and training cohort size. The benchmark pipeline and code are released publicly.

Significance. If the comparison is sound, AdaBrain-Bench would be a valuable community resource: it is broader than existing EEG benchmarks in task coverage, explicitly contrasts foundation models with traditional baselines, and ships a reproducible pipeline. The inclusion of three transfer settings and a transferability metric is a useful step toward standardized evaluation in a fast-moving field. However, the central quantitative claims rest on two assumptions that the manuscript does not adequately support: that evaluations on datasets included in a model's pretraining still measure cross-subject generalization, and that macro-averaging heterogeneous metrics yields a meaningful summary. These issues are fixable but require re-analysis, so the current version overstates the evidence for the headline conclusion.

major comments (4)
  1. [§2, Tables 2 and 7] The evaluation is partially circular at the dataset level. LaBraM's pretraining corpus includes 'SEED series' and 'Siena'; CBraMod's includes TUAB and TUEV; BIOT's includes SHHS, TUAB, and TUEV; and EEGPT's includes SEED. All of these datasets also appear in the AdaBrain-Bench evaluation (Table 6). The cross-subject protocol excludes test subjects only from fine-tuning, not from pretraining, so the reported gains on these datasets may reflect in-domain familiarity rather than generic transfer. The headline macro-average (LaBraM 64.61% and CBraMod 62.66% vs. Conformer 58.12%) and the transfer scores in Table 4 are therefore not clean evidence for the claim that foundation models 'show great potential for enhancing cross-subject generalization performance.' Please re-quantify the comparison using only datasets that are absent from each model's pretraining corpus, or otherwise explicitly control for pretraining overlap.
  2. [§2, Table 2 macro-average] The macro-average that drives the central claim averages over incommensurable metrics: balanced accuracy, AUROC, AUC-PR, Pearson correlation (reported on a 0–100 scale), R², 2-way retrieval accuracy, and Top-5 accuracy. These quantities have different scales, ranges, and interpretations (R² can be negative), so a single macro-average is not a principled summary of 'overall' performance. The numbers 64.61% and 58.12% are thus not directly interpretable as accuracy-like scores. Please report per-domain aggregates (e.g., cognitive state, human augmentation, clinical monitoring) or standardize metrics before averaging, and avoid presenting the macro-average as the primary evidence.
  3. [Tables 2–5] All benchmark tables report point estimates without variance or statistical significance. The few-shot results are said to be averaged over three runs, but no standard deviations are given, and the cross-subject results appear to come from a single split. Several reported differences are small (e.g., Sleep-EDF: 69.47% for the best foundation model vs. 69.55% for the best traditional model), so the claim that foundation models 'consistently outperform' baselines is not statistically supported. Please report mean ± std over multiple seeds and, where relevant, over subject splits, and add a significance test or confidence intervals for the key comparisons.
  4. [§4, Eq. (1)] The transfer score mixes two different denominators: Poracle is set to 1 for cross-subject transfer but to full-data performance for few-shot transfer, and λ = 0.5 is chosen without sensitivity analysis. Consequently, transfer scores are not comparable across the two settings, and the 'Average' column in Table 4 averages quantities with different upper bounds. More importantly, when the downstream dataset appeared in pretraining, the comparison Pprt − Pscr does not isolate transfer from domain familiarity; it credits in-domain pretraining as if it were cross-domain transfer. Please clarify the interpretation of TS or redefine it so that it measures transfer to genuinely unseen data.
minor comments (6)
  1. [§4, Evaluation Settings] The few-shot sampling ratios are inconsistent between the Results section (r = [0.02, 0.05, 0.1, 0.3, 0.5]) and the Methods section (k ranging from [0.02, 0.05, 0.3, 0.5]); the value 0.1 and the variable name should be unified.
  2. [§2, Impact of Number of Training Subjects] The text states that HMC and SHHS include 120 and 263 subjects, respectively, but Table 6 lists 151 and 329 subjects; these numbers should be reconciled.
  3. [§2, Impact of Pretraining; Table 7] The pretraining duration for CBraMod is given as approximately 27,000 hours in the text but 9,246 hours in Table 7, and EEGPT is listed as 246 hours in the text but 198 hours in Table 7; these inconsistencies should be corrected.
  4. [§4, Other Experimental Details] The learning-rate grid is written as [1e−3, 5e−4, 1e−4, 5e−4, 1e−5], which contains a duplicate 5e−4 and omits the intended 1e−2; this is likely a typo.
  5. [Table 5] The dataset name 'EDMAT' in Table 5 should be 'EEGMAT' for consistency with the rest of the paper.
  6. [Table 2, SEED-VIG] The Corr values for SEED-VIG (e.g., 58.34) appear to be Pearson correlation multiplied by 100, but the scaling is not stated in the table caption or the metric definition in Section 4.

Circularity Check

1 steps flagged · score 4.0 of 10

Cross-subject gains partly reflect pretraining on benchmark datasets; no derivational circularity in the transfer-score definition, but the headline claim is confounded by pretraining/evaluation overlap.

  1. other [Section 2 (cross-subject results) combined with Section 4 (Cross-Subject Transfer Setting definition) and Tables 6-7 (dataset overlap)]
    "Cross-Subject Transfer Setting. This setting trains or fine-tunes models on EEG data from a group of subjects and tests them on new subjects whose data is explicitly excluded from training. Table 7 pretraining data: 'BCI-IV-1, Emobrain, SPIS,Grasp and Lift EEG, Inria P300,SEED series, Siena,TUEP, TUSZ & others' (LaBraM), 'PREST, SHHS, CHB-MIT,IIIC Seizure, TUAB & TUEV' (BIOT), 'PhysioNet-MI, HGD, TSU,SEED & M3CV' (EEGPT), 'TUAB, TUARTUEP, TUEVTUSE & TUSLF' (CBraMod)."

    The cross-subject protocol defines 'unseen' subjects only with respect to downstream fine-tuning, not self-supervised pretraining. Several evaluation datasets in Table 6 appear verbatim in Table 7 pretraining corpora: LaBraM pretrained on SEED series and Siena; CBraMod on TUAB/TUEV; BIOT on SHHS/TUAB/TUEV; EEGPT on SEED. For these datasets the test cohort's data (or at least the same dataset statistics and task structure) was already seen by the pretrained model, so the reported cross-subject scores and transfer scores credit in-domain pretraining as generic cross-subject transfer.

full rationale

This is a benchmark paper, not a mathematical derivation; the only defined quantity (transfer score, Eq. 1) is transparently a relative improvement over a from-scratch counterpart and is not itself circular. However, the central empirical claim is weakened by a concrete data-overlap confound: the cross-subject setting excludes test subjects only from fine-tuning, while LaBraM, CBraMod, BIOT, and EEGPT were pretrained on several of the same datasets used for evaluation (SEED, SEED-IV, TUEV, TUAB, Siena, SHHS). For those datasets, the 'unseen subjects' were not unseen to the pretrained model, so the headline macro-averages (LaBraM 64.61%, CBraMod 62.66% vs. Conformer 58.12%) mix genuine transfer with in-domain familiarity. The paper never discloses this overlap or re-quantifies results on truly unseen tasks. Because the benchmark itself is a useful, reproducible artifact and several evaluated datasets are genuinely absent from each model's pretraining, the circularity is partial rather than total; no self-citation chain or fitted-parameter issue is present.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central empirical claims rest on the representativeness of the 13-dataset selection, the fidelity of the public checkpoints, the neutrality of the channel-projection adaptation, and the validity of held-out-subject evaluation on datasets already present in pretraining corpora. The transfer score adds two hand-chosen quantities (lambda and the oracle reference) that shape the reported transferability numbers. No new physical entities are introduced.

free parameters (2)
  • λ in transfer score (Eq. 1) = 0.5
    Hand-chosen weighting between relative gain and relative improvement toward the oracle; directly sets every reported transfer score.
  • Poracle for few-shot transfer score = full-data performance (per model, per dataset)
    The 'oracle' upper bound in Eq. (1) is defined as the full-data performance, a per-model, per-dataset choice that determines the second term of the transfer score.
assumptions (4)
  • domain assumption The 13 selected datasets and 7 tasks are representative of the key non-invasive BCI applications.
    The paper motivates the benchmark by this representativeness (Table 6), and the headline conclusions about foundation-model utility generalize only if this selection is representative.
  • domain assumption Publicly released checkpoints of BIOT, EEGPT, LaBraM, and CBraMod faithfully match the models described in their papers.
    All evaluations depend on using the official pretrained weights as-is; the paper never verifies checkpoint contents or versions.
  • ad hoc to paper A 1x1 channel-wise convolutional projection for BIOT and EEGPT on channel-mismatched datasets does not bias the comparison against those models.
    This projection is introduced in Methods (Other Experimental Details) and is trained from scratch, so under few-shot settings it might disadvantage models that need it; this is a paper-specific adaptation not part of the original foundation models.
  • ad hoc to paper Held-out-subject splits within datasets seen during pretraining still measure generalization.
    The benchmark evaluates on datasets that appear in pretraining corpora (e.g., LaBraM on SEED series, Siena, TUEV; CBraMod on TUEG; BIOT on SHHS, TUAB, TUEV). The paper implicitly assumes the held-out-subject protocol removes the benefit of pretraining on the same distribution.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AdaBrain-Bench: Benchmarking Brain Foundation Models for Brain-Computer Interface Applications." pith.science (2026). https://pith.science/paper/57SEPODV

@misc{pith2026250709882,
  author       = {Pith},
  title        = {Pith review of: AdaBrain-Bench: Benchmarking Brain Foundation Models for Brain-Computer Interface Applications},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/57SEPODV}},
  note         = {Machine review of arXiv:2507.09882}
}
read the original abstract

Non-invasive Brain-Computer Interfaces (BCI) offer a safe and accessible means of connecting the human brain to external devices, with broad applications in home and clinical settings to enhance human capabilities. However, the high noise level and limited task-specific data in non-invasive signals constrain decoding capabilities. Recently, the adoption of self-supervised pre-training is transforming the landscape of non-invasive BCI research, enabling the development of brain foundation models to capture generic neural representations from large-scale unlabeled electroencephalography (EEG) signals with substantial noises. However, despite these advances, the field currently lacks comprehensive, practical and extensible benchmarks to assess the utility of the public foundation models across diverse BCI tasks, hindering their widespread adoption. To address this challenge, we present AdaBrain-Bench, a large-scale standardized benchmark to systematically evaluate brain foundation models in widespread non-invasive BCI tasks. AdaBrain-Bench encompasses a diverse collection of representative BCI decoding datasets spanning 7 key applications. It introduces a streamlined task adaptation pipeline integrated with multi-dimensional evaluation metrics and a set of adaptation tools. The benchmark delivers an inclusive framework for assessing generalizability of brain foundation models across key transfer settings, including cross-subject, multi-subject, and few-shot scenarios. We leverage AdaBrain-Bench to evaluate a suite of publicly available brain foundation models and offer insights into practices for selecting appropriate models in various scenarios. We make our benchmark pipeline available to enable reproducible research and external use, offering a continuously evolving platform to foster progress toward robust and generalized neural decoding solutions.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 5 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. BrainBench: Benchmarking Large Language Models for Comprehensive EEG Understanding

    cs.AI 2026-08 conditional novelty 7.0 of 10

    A unified benchmark of 172 EEG analysis tasks shows that LLMs handle well-specified analyses better than long multi-step workflows, and that structured agent execution usually beats autonomous code generation.

  2. What EEG Foundation Models Encode: Dataset Identity and a Negative-Control Suite for Clinical Benchmarks

    cs.LG 2026-07 conditional novelty 6.5 of 10

    Pretrained EEG foundation models do not transfer disease information across populations in this benchmark: classical features beat them on an external Korean cohort, dataset identity is decodable at ceiling, and only ...

  3. NeuroOnline: Bridging Pretraining and Online Adaptation for EEG Foundation Models

    cs.LG 2026-07 conditional novelty 5.5 of 10

    NeuroOnline adapts EEG foundation models online via multi-view consistency and context-prompt modulation, outperforming static fine-tuning under distribution shifts on motor-imagery and emotion benchmarks.

  4. WorldWeaver: Generating Long-Horizon Video Worlds via Rich Perception

    cs.CV 2025-08 unverdicted novelty 5.0 of 10

    WorldWeaver reduces temporal drift in long-horizon video generation by jointly modeling RGB and depth perceptual conditions with segmented noise scheduling.

  5. Foundation Models for Cross-Domain EEG Analysis Application: A Survey

    cs.HC 2025-08 conditional novelty 4.0 of 10

    A survey that organizes EEG foundation-model research into five output-modality categories: native EEG, text, vision, audio, and multimodal fusion, with a claim to be the first such comprehensive taxonomy.

Reference graph

Works this paper leans on

72 extracted references · 68 canonical work pages · cited by 5 Pith papers

  1. [1]

    IEEE Transactions on Biomedical Engineering 52(9), 1541–1548 (2005)

    Lemm, S., Blankertz, B., Curio, G., Muller, K.-R.: Spatio-spectral filters for improving the classification of single trial eeg. IEEE Transactions on Biomedical Engineering 52(9), 1541–1548 (2005)

    work page 2005

  2. [2]

    IEEE Transactions on Pattern Analysis and Machine Intelligence 33(3), 433–445 (2010)

    Cecotti, H., Graser, A.: Convolutional neural networks for p300 detection with application to brain-computer interfaces. IEEE Transactions on Pattern Analysis and Machine Intelligence 33(3), 433–445 (2010)

    work page 2010

  3. [3]

    IEEE Transactions on Systems, man, and cybernetics: systems 46(6), 843–854 (2015)

    He, L., Hu, D., Wan, M., Wen, Y., Von Deneen, K.M., Zhou, M.: Common bayesian network for classification of eeg-based multiclass motor imagery bci. IEEE Transactions on Systems, man, and cybernetics: systems 46(6), 843–854 (2015)

    work page 2015

  4. [4]

    Science 372(6544), 831–836 (2021)

    Flesher, S.N., Downey, J.E., Weiss, J.M., Hughes, C.L., Herrera, A.J., Tyler- Kabara, E.C., Boninger, M.L., Collinger, J.L., Gaunt, R.A.: A brain-computer interface that evokes tactile sensations improves robotic arm control. Science 372(6544), 831–836 (2021)

    work page 2021

  5. [5]

    Ding, Y., Udompanyawit, C., Zhang, Y., He, B.: Eeg-based brain-computer interface enables real-time robotic hand control at individual finger level. Nat. Commun. 16(1), 1–20 (2025)

    work page 2025

  6. [6]

    Nature 568(7753), 493–498 (2019)

    Anumanchipalli, G.K., Chartier, J., Chang, E.F.: Speech synthesis from neural decoding of spoken sentences. Nature 568(7753), 493–498 (2019)

    work page 2019

  7. [7]

    : A high-performance speech neuroprosthesis

    Willett, F.R., Kunz, E.M., Fan, C., Avansino, D.T., Wilson, G.H., Choi, E.Y., Kamdar, F., Glasser, M.F., Hochberg, L.R., Druckmann, S., et al. : A high-performance speech neuroprosthesis. Nature 620(7976), 1031–1036 (2023) 22

    work page 2023

  8. [8]

    Tang, J., LeBel, A., Jain, S., Huth, A.G.: Semantic reconstruction of continu- ous language from non-invasive brain recordings. Nat. Neurosci. 26(5), 858–866 (2023)

    work page 2023

Show all 72 references
  1. [9]

    Chen, X., Wang, R., Khalilian-Gourtani, A., Yu, L., Dugan, P., Friedman, D., Doyle, W., Devinsky, O., Wang, Y., Flinker, A.: A neural speech decoding frame- work leveraging deep learning and speech synthesis. Nat. Mach. Intell. 6(4), 467–480 (2024)

    2024

  2. [10]

    Nature, 1–8 (2025)

    Wairagkar, M., Card, N.S., Singer-Clark, T., Hou, X., Iacobacci, C., Miller, L.M., Hochberg, L.R., Brandman, D.M., Stavisky, S.D.: An instantaneous voice- synthesis neuroprosthesis. Nature, 1–8 (2025)

    2025

  3. [11]

    Neurocomputing 223, 26–44 (2017)

    Ramadan, R.A., Vasilakos, A.V.: Brain computer interface: control signals review. Neurocomputing 223, 26–44 (2017)

    2017

  4. [12]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering 25(4), 392–401 (2016)

    Ang, K.K., Guan, C.: Eeg-based strategies to detect motor imagery for con- trol and rehabilitation. IEEE Transactions on Neural Systems and Rehabilitation Engineering 25(4), 392–401 (2016)

    2016

  5. [13]

    IEEE Journal of Biomedical and Health Informatics 28(2), 765–776 (2023)

    Han, J., Gu, X., Yang, G.-Z., Lo, B.: Noise-factorized disentangled representa- tion learning for generalizable motor imagery eeg classification. IEEE Journal of Biomedical and Health Informatics 28(2), 765–776 (2023)

    2023

  6. [14]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering27(6), 1170–1180 (2019)

    Li, Y., Zhang, X., Zhang, B., Lei, M., Cui, W., Guo, Y.: A channel-projection mixed-scale convolutional neural network for motor imagery eeg decoding. IEEE Transactions on Neural Systems and Rehabilitation Engineering27(6), 1170–1180 (2019)

    2019

  7. [15]

    Nature Communications 16, 5401 (2025)

    Ding, Y., Udompanyawit, C., Zhang, Y., He, B.: Eeg-based brain-computer inter- face enables real-time robotic hand control at individual finger level. Nature Communications 16, 5401 (2025)

    2025

  8. [16]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering 25(11), 1998–2008 (2017)

    Supratak, A., Dong, H., Wu, C., Guo, Y.: Deepsleepnet: A model for automatic sleep stage scoring based on raw single-channel eeg. IEEE Transactions on Neural Systems and Rehabilitation Engineering 25(11), 1998–2008 (2017)

    2017

  9. [17]

    Computers in Biology and Medicine 100, 270–278 (2018)

    Acharya, U.R., Oh, S.L., Hagiwara, Y., Tan, J.H., Adeli, H.: Deep convolutional neural network for the automated detection and diagnosis of seizure using eeg signals. Computers in Biology and Medicine 100, 270–278 (2018)

    2018

  10. [18]

    IEEE Transactions on Affective Computing 11(3), 532–541 (2018)

    Song, T., Zheng, W., Song, P., Cui, Z.: Eeg emotion recognition using dynamical graph convolutional neural networks. IEEE Transactions on Affective Computing 11(3), 532–541 (2018)

    2018

  11. [19]

    Journal of Neural Engineering 9(4), 045008 (2012)

    Brouwer, A.-M., Hogervorst, M.A., Van Erp, J.B., Heffelaar, T., Zimmerman, P.H., Oostenveld, R.: Estimating workload using eeg spectral power and erps in 23 the n-back task. Journal of Neural Engineering 9(4), 045008 (2012)

    2012

  12. [20]

    In: Advances in Neural Information Processing Systems, vol

    Li, D., Wei, C., Li, S., Zou, J., Qin, H., Liu, Q.: Visual decoding and reconstruction via eeg embeddings with guided diffusion. In: Advances in Neural Information Processing Systems, vol. 37, pp. 102822–102864 (2024)

    2024

  13. [21]

    In: Advances in Neural Information Processing Systems, vol

    Yang, C., Westover, M., Sun, J.: Biot: Biosignal transformer for cross-data learn- ing in the wild. In: Advances in Neural Information Processing Systems, vol. 36, pp. 78240–78260 (2023)

    2023

  14. [22]

    In: Advances in Neural Information Processing Systems, vol

    Wang, G., Liu, W., He, Y., Xu, C., Ma, L., Li, H.: Eegpt: Pretrained transformer for universal and reliable representation of eeg signals. In: Advances in Neural Information Processing Systems, vol. 37, pp. 39249–39280 (2024)

    2024

  15. [23]

    In: International Conference on Learning Representations (2024)

    Jiang, W., Zhao, L., Lu, B.: Large brain model for learning generic representa- tions with tremendous eeg data in bci. In: International Conference on Learning Representations (2024)

    2024

  16. [24]

    In: International Conference on Learning Representations (2025)

    Wang, J., Zhao, S., Luo, Z., Zhou, Y., Jiang, H., Li, S., Li, T., Pan, G.: CBramod: A criss-cross brain foundation model for EEG decoding. In: International Conference on Learning Representations (2025)

    2025

  17. [25]

    Journal of neural engineering 15(5), 056013 (2018)

    Lawhern, V.J., Solon, A.J., Waytowich, N.R., Gordon, S.M., Hung, C.P., Lance, B.J.: Eegnet: a compact convolutional neural network for eeg-based brain– computer interfaces. Journal of neural engineering 15(5), 056013 (2018)

    2018

  18. [26]

    NeuroImage 276, 120209 (2023)

    Miao, Z., Zhao, M., Zhang, X., Ming, D.: Lmda-net: A lightweight multi- dimensional attention network for general eeg-based brain-computer interfaces and interpretability. NeuroImage 276, 120209 (2023)

    2023

  19. [27]

    arXiv preprint arXiv:2106.11170 (2021)

    Song, Y., Jia, X., Yang, L., Xie, L.: Transformer-based spatial-temporal feature learning for eeg decoding. arXiv preprint arXiv:2106.11170 (2021)

    2021 arXiv

  20. [28]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering 31, 710–719 (2022)

    Song, Y., Zheng, Q., Liu, B., Gao, X.: Eeg conformer: Convolutional transformer for eeg decoding and visualization. IEEE Transactions on Neural Systems and Rehabilitation Engineering 31, 710–719 (2022)

    2022

  21. [29]

    IEEE Transactions on Neural Networks and Learning Systems 29(11), 5619–5629 (2018)

    Sakhavi, S., Guan, C., Yan, S.: Learning temporal information for brain-computer interface using convolutional neural networks. IEEE Transactions on Neural Networks and Learning Systems 29(11), 5619–5629 (2018)

    2018

  22. [30]

    IEEE Transactions on Neural Networks and Learning Systems30(9), 2755–2763 (2019)

    Gao, Z., Wang, X., Yang, Y., Mu, C., Cai, Q., Dang, W., Zuo, S.: Eeg-based spatio–temporal convolutional neural network for driver fatigue evaluation. IEEE Transactions on Neural Networks and Learning Systems30(9), 2755–2763 (2019)

    2019

  23. [31]

    IEEE Transactions on Neural Systems and Rehabilitation 24 Engineering 26(11), 2086–2095 (2018)

    Wang, P., Jiang, A., Liu, X., Shang, J., Zhang, L.: Lstm-based eeg classification in motor imagery tasks. IEEE Transactions on Neural Systems and Rehabilitation 24 Engineering 26(11), 2086–2095 (2018)

    2018

  24. [32]

    Computers in Biology and Medicine 99, 24–37 (2018)

    Tsiouris, K.M., Pezoulas, V.C., Zervakis, M., Konitsiotis, S., Koutsouris, D.D., Fotiadis, D.I.: A long short-term memory deep learning network for the prediction of epileptic seizures using eeg signals. Computers in Biology and Medicine 99, 24–37 (2018)

    2018

  25. [33]

    IEEE Transactions on Affective Computing 13(3), 1528–1540 (2020)

    Du, X., Ma, C., Zhang, G., Li, J., Lai, Y.-K., Zhao, G., Deng, X., Liu, Y.-J., Wang, H.: An efficient lstm network for emotion recognition from multichannel eeg signals. IEEE Transactions on Affective Computing 13(3), 1528–1540 (2020)

    2020

  26. [34]

    IEEE Transactions on Biomedical Engineering 69(8), 2456–2467 (2022)

    Phan, H., Mikkelsen, K., Ch´ en, O.Y., Koch, P., Mertins, A., De Vos, M.: Sleeptransformer: Automatic sleep staging with interpretability and uncertainty quantification. IEEE Transactions on Biomedical Engineering 69(8), 2456–2467 (2022)

    2022

  27. [35]

    Human Brain Mapping 38(11), 5391–5420 (2017)

    Schirrmeister, R.T., Springenberg, J.T., Fiederer, L.D.J., Glasstetter, M., Eggensperger, K., Tangermann, M., Hutter, F., Burgard, W., Ball, T.: Deep learn- ing with convolutional neural networks for eeg decoding and visualization. Human Brain Mapping 38(11), 5391–5420 (2017)

    2017

  28. [36]

    D´ efossez, A., Caucheteux, C., Rapin, J., Kabeli, O., King, J.-R.: Decoding speech perception from non-invasive brain recordings. Nat. Mach. Intell. 5(10), 1097– 1107 (2023)

    2023

  29. [37]

    IEEE Transactions on Neural Networks and Learning Systems 32(2), 535–545 (2020)

    Zhao, H., Zheng, Q., Ma, K., Li, H., Zheng, Y.: Deep representation-based domain adaptation for nonstationary eeg classification. IEEE Transactions on Neural Networks and Learning Systems 32(2), 535–545 (2020)

    2020

  30. [38]

    Biomedical Signal Processing and Control 63, 102144 (2021)

    Zhang, R., Zong, Q., Dou, L., Zhao, X., Tang, Y., Li, Z.: Hybrid deep neural net- work using transfer learning for eeg motor imagery decoding. Biomedical Signal Processing and Control 63, 102144 (2021)

    2021

  31. [39]

    In: International Conference on Learning Representations Workshop (2025)

    Lee, N., Bakas, S., Barmpas, K., Panagakis, Y., Adamos, D., Laskaris, N., Zafeiriou, S.: Assessing the capabilities of large brainwave foundation models. In: International Conference on Learning Representations Workshop (2025)

    2025

  32. [40]

    arXiv preprint arXiv:2410.09767 (2024)

    Liu, H., Yang, S., Zhang, Y., Wang, M., Gong, F., Xie, C., Liu, G., Liu, Z., Liu, Y.-J., Lu, B.-L., et al.: Libeer: A comprehensive benchmark and algorithm library for eeg-based emotion recognition. arXiv preprint arXiv:2410.09767 (2024)

    2024 arXiv

  33. [41]

    Epilepsia (2024) 25

    Dan, J., Pale, U., Amirshahi, A., Cappelletti, W., Ingolfsson, T.M., Wang, X., Cossettini, A., Bernini, A., Benini, L., Beniczky, S., et al.: Szcore: Seizure community open-source research evaluation framework for the validation of electroencephalography-based automated seizur...

    2024

  34. [42]

    arXiv preprint arXiv:2404.15319 (2024)

    Chevallier, S., Carrara, I., Aristimunha, B., Guetschel, P., Sedlar, S., Lopes, B., Velut, S., Khazem, S., Moreau, T.: The largest eeg-based bci reproducibility study for open science: the moabb benchmark. arXiv preprint arXiv:2404.15319 (2024)

    2024 arXiv

  35. [43]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Li, B., Ge, Y., Ge, Y., Wang, G., Wang, R., Zhang, R., Shan, Y.: Seed- bench: Benchmarking multimodal large language models. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 13299–13308 (2024)

    2024

  36. [44]

    Transactions of the Association for Computational Linguistics 12, 39–57 (2024)

    Zhang, T., Ladhak, F., Durmus, E., Liang, P., McKeown, K., Hashimoto, T.B.: Benchmarking large language models for news summarization. Transactions of the Association for Computational Linguistics 12, 39–57 (2024)

    2024

  37. [45]

    : Cambrian-1: A fully open, vision-centric exploration of multimodal llms

    Tong, P., Brown, E., Wu, P., Woo, S., IYER, A.J.V., Akula, S.C., Yang, S., Yang, J., Middepogu, M., Wang, Z., et al. : Cambrian-1: A fully open, vision-centric exploration of multimodal llms. Advances in Neural Information Processing Systems 37, 87310–87356 (2024)

    2024

  38. [46]

    IEEE Transactions on Autonomous Mental Development 7(3), 162–175 (2015)

    Zheng, W., Lu, B.: Investigating critical frequency bands and channels for eeg- based emotion recognition with deep neural networks. IEEE Transactions on Autonomous Mental Development 7(3), 162–175 (2015)

    2015

  39. [47]

    IEEE Transactions on Cybernetics 49(3), 1110–1122 (2018)

    Zheng, W.-L., Liu, W., Lu, Y., Lu, B.-L., Cichocki, A.: Emotionmeter: A multimodal framework for recognizing human emotions. IEEE Transactions on Cybernetics 49(3), 1110–1122 (2018)

    2018

  40. [48]

    Data 4(1), 14 (2019)

    Zyma, I., Tukaev, S., Seleznov, I., Kiyono, K., Popov, A., Chernykh, M., Shpenkov, O.: Electroencephalograms during mental arithmetic task perfor- mance. Data 4(1), 14 (2019)

    2019

  41. [49]

    Circulation 101(23), 215–220 (2000)

    Goldberger, A.L., Amaral, L.A., Glass, L., Hausdorff, J.M., Ivanov, P.C., Mark, R.G., Mietus, J.E., Moody, G.B., Peng, C.-K., Stanley, H.E.: Physiobank, phys- iotoolkit, and physionet: components of a new research resource for complex physiologic signals. Circulation 101(23), ...

    2000

  42. [50]

    Journal of Neural Engineering 14(2), 026017 (2017)

    Zheng, W.-L., Lu, B.-L.: A multimodal approach to estimating vigilance using eeg and forehead eog. Journal of Neural Engineering 14(2), 026017 (2017)

    2017

  43. [51]

    : Review of the bci competition iv

    Tangermann, M., M¨ uller, K.-R., Aertsen, A., Birbaumer, N., Braun, C., Brunner, C., Leeb, R., Mehring, C., Miller, K.J., M¨ uller-Putz, G.R., et al. : Review of the bci competition iv. Frontiers in neuroscience 6, 55 (2012)

    2012

  44. [52]

    Scientific Data 9(1), 531 (2022)

    Ma, J., Yang, B., Qiu, W., Li, Y., Gao, S., Xia, X.: A large eeg dataset for study- ing cross-session variability in motor imagery brain-computer interface. Scientific Data 9(1), 531 (2022)

    2022

  45. [53]

    NeuroImage 264, 119754 (2022)

    Gifford, A.T., Dwivedi, K., Roig, G., Cichy, R.M.: A large and rich eeg dataset 26 for modeling human visual object recognition. NeuroImage 264, 119754 (2022)

    2022

  46. [54]

    Scientific Data 9(1), 3 (2022)

    Grootswagers, T., Zhou, I., Robinson, A.K., Hebart, M.N., Carlson, T.A.: Human eeg recordings for 1,854 concepts presented in rapid serial visual presentation streams. Scientific Data 9(1), 3 (2022)

    2022

  47. [55]

    In: Conference on Neural Information Processing Systems (2024)

    Li, D., Wei, C., Li, S., Zou, J., Liu, Q.: Visual decoding and reconstruction via eeg embeddings with guided diffusion. In: Conference on Neural Information Processing Systems (2024)

    2024

  48. [56]

    Frontiers in Neuroscience 10, 196 (2016)

    Obeid, I., Picone, J.: The temple university hospital eeg data corpus. Frontiers in Neuroscience 10, 196 (2016)

    2016

  49. [57]

    Processes 8(7) (2020) https://doi.org/ 10.3390/pr8070846

    Detti, P., Vatti, G., Lara, G.: Eeg synchronization analysis for seizure prediction: A study on data of noninvasive recordings. Processes 8(7) (2020) https://doi.org/ 10.3390/pr8070846

    2020 doi

  50. [58]

    PloS one 16(8), 0256111 (2021)

    Alvarez-Estevez, D., Rijsman, R.M.: Inter-database validation of a deep learning approach for automatic sleep scoring. PloS one 16(8), 0256111 (2021)

    2021

  51. [59]

    Journal of the American Medical Informatics Association 25(10), 1351–1358 (2018)

    Zhang, G.-Q., Cui, L., Mueller, R., Tao, S., Kim, M., Rueschman, M., Mariani, S., Mobley, D., Redline, S.: The national sleep research resource: towards a sleep data commons. Journal of the American Medical Informatics Association 25(10), 1351–1358 (2018)

    2018

  52. [60]

    : The sleep heart health study: design, rationale, and methods

    Quan, S.F., Howard, B.V., Iber, C., Kiley, J.P., Nieto, F.J., O’Connor, G.T., Rapoport, D.M., Redline, S., Robbins, J., Samet, J.M., et al. : The sleep heart health study: design, rationale, and methods. Sleep 20(12), 1077–1085 (1997)

    1997

  53. [61]

    IEEE Transactions on Biomedical Engineering 47(9), 1185–1194 (2000)

    Kemp, B., Zwinderman, A.H., Tuk, B., Kamphuisen, H.A., Oberye, J.J.: Analysis of a sleep-dependent neuronal feedback loop: the slow-wave microcontinuity of the eeg. IEEE Transactions on Biomedical Engineering 47(9), 1185–1194 (2000)

    2000

  54. [62]

    Rules, Terminology and Technical Specifications, Darien, Illinois, American Academy of Sleep Medicine 176, 7 (2012)

    Berry, R.B., Brooks, R., Gamaldo, C.E., Harding, S.M., Marcus, C., Vaughn, B.V., et al.: The aasm manual for the scoring of sleep and associated events. Rules, Terminology and Technical Specifications, Darien, Illinois, American Academy of Sleep Medicine 176, 7 (2012)

    2012

  55. [63]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering 29, 809–818 (2021)

    Eldele, E., Chen, Z., Liu, C., Wu, M., Kwoh, C.-K., Li, X., Guan, C.: An attention- based deep learning approach for sleep stage classification with single-channel eeg. IEEE Transactions on Neural Systems and Rehabilitation Engineering 29, 809–818 (2021)

    2021

  56. [64]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering 30, 1992–2002 (2022) 27

    Chen, Y., Yang, R., Huang, M., Wang, Z., Liu, X.: Single-source to single-target cross-subject motor imagery classification based on multisubdomain adaptation network. IEEE Transactions on Neural Systems and Rehabilitation Engineering 30, 1992–2002 (2022) 27

    2022

  57. [65]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering 16(5), 432–441 (2008)

    Bigdely-Shamlo, N., Vankov, A., Ramirez, R.R., Makeig, S.: Brain activity-based image classification from rapid serial visual presentation. IEEE Transactions on Neural Systems and Rehabilitation Engineering 16(5), 432–441 (2008)

    2008

  58. [66]

    Neurocomputing 74(12-13), 2041–2051 (2011)

    Huang, Y., Erdogmus, D., Pavel, M., Mathan, S., Hild II, K.E.: A frame- work for rapid visual image search using single-trial brain evoked responses. Neurocomputing 74(12-13), 2041–2051 (2011)

    2011

  59. [67]

    Neuropsychobiology 66(4), 201–220 (2012)

    Jobert, M., Wilson, F.J., Ruigt, G.S., Brunovsky, M., Prichep, L.S., Drinken- burg, W.H., Committee, I.P.-E.G.: Guidelines for the recording and evaluation of pharmaco-eeg data in man: the international pharmaco-eeg society (ipeg). Neuropsychobiology 66(4), 201–220 (2012)

    2012

  60. [68]

    IEEE Transactions on Neural Systems and Rehabilitation Engineering 26(2), 371–382 (2017)

    Saha, S., Ahmed, K.I.U., Mostafa, R., Hadjileontiadis, L., Khandoker, A.: Evidence of variabilities in eeg dynamics during motor imagery-based mul- ticlass brain–computer interface. IEEE Transactions on Neural Systems and Rehabilitation Engineering 26(2), 371–382 (2017)

    2017

  61. [69]

    IEEE Transactions on Cybernetics 55(5), 2038–2051 (2025)

    Chen, C.P., Chen, B., Zhang, T.: Adamgraph: Adaptive attention-modulated graph network for eeg emotion recognition. IEEE Transactions on Cybernetics 55(5), 2038–2051 (2025)

    2025

  62. [70]

    In: International Conference on Machine Learning, pp

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., et al.: Learning transferable visual models from natural language supervision. In: International Conference on Machine Learning, pp. 8748–8763 (2021)

    2021

  63. [71]

    In: Proceedings of the IEEE/CVF Conference on Computer Cision and Pattern Recognition, pp

    Zhou, K., Yang, J., Loy, C.C., Liu, Z.: Conditional prompt learning for vision- language models. In: Proceedings of the IEEE/CVF Conference on Computer Cision and Pattern Recognition, pp. 16816–16825 (2022)

    2022

  64. [72]

    2661–2671 (2019) 28

    Kornblith, S., Shlens, J., Le, Q.V.: Do better imagenet models transfer better? In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 2661–2671 (2019) 28

    2019

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.