Pith. sign in

REVIEW 3 major objections 5 minor 42 references

NeuroOnline: Bridging Pretraining and Online Adaptation for EEG Foundation Models

T0 review · 3 major / 5 minor · reviewed 2026-07-11 · grok-4.5

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

Pith's one-line read EEG foundation models can keep improving after deployment if they align multi-view representations and modulate them with a learnable context prompt.

desk verdict Solid empirical recipe for online adaptation of EEG foundation models; gains look real but the evaluation protocol is under-specified and the online baseline is thin. read the letter →

arxiv 2607.03925 v1 pith:OGETJPTN submitted 2026-07-04 cs.LG

Weibin Li , Wendu Li , Yushan You , Chen Wei , Quanying Liu This is my paper · ORCID
classification cs.LG
keywords EEGfoundationmodelsonlineadaptationmulti-viewconsistencycontextpromptdistributionshiftmotorimageryemotionrecognitiontest-time
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

Most EEG foundation models are frozen after pretraining, so they cannot fix the mismatch between their original training objective and a real decoding task, and they cannot track the non-stationary shifts that appear once a headset is worn. NeuroOnline replaces that static paradigm with continuous online updates driven by two mechanisms: multi-view consistency (forcing the model to treat temporal and frequency-masked versions of each new sample the same way) and context-aware modulation (a learnable prompt that scales and shifts features via cross-attention). Across four foundation backbones and six motor-imagery and emotion-recognition benchmarks, the joint method beats both strong offline fine-tunes and a plain online fine-tune baseline, with the largest gains on multi-class problems. The result matters because it shows that foundation-model performance need not be fixed at deployment; modest online supervision plus these two adaptation modules can keep accuracy rising as brain signals drift.

What carries the argument

NeuroOnline: multi-view consistency learning (aligning modulated embeddings of an anchor EEG segment with temporally and frequency-masked views) plus context-aware representation modulation (a learnable prompt that produces scale and shift parameters via cross-attention).

What would settle it

Run the identical online protocol on any of the six benchmarks but withhold labels after the first few hundred samples; if performance then collapses to or below the plain online baseline, the claim that the two mechanisms enable sustained adaptation fails.

Watch

Extended reading notes

Core claim

The paper establishes that jointly enforcing multi-view consistency and context-aware representation modulation lets pretrained EEG foundation models adapt continuously in online streams, closing the pretraining–task gap and handling distribution shift; the resulting NeuroOnline framework statistically significantly outperforms both offline fine-tuning and naïve online updating across multiple backbones and six EEG benchmarks.

Load-bearing premise

Every online EEG sample must arrive with its true task label so the supervised classification loss can be computed and the model updated immediately.

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

3 major / 5 minor

Summary. The paper proposes NeuroOnline, a framework for continual online adaptation of pretrained EEG foundation models. It combines multi-view consistency learning (MCL), which aligns representations of temporally/frequency-masked views of each online sample, with context-aware representation modulation (CRM), which uses a learnable prompt and cross-attention to produce input-dependent scale/shift parameters. The overall loss is supervised classification plus a consistency term (Eq. 4). Experiments on six public EEG benchmarks (motor imagery and emotion) with four foundation-model backbones report statistically significant gains over offline fine-tuning variants and an ablated online baseline (NeuroOnline.a), supported by ablations and a λ sensitivity sweep.

Significance. Non-stationarity and pretraining–task mismatch are genuine obstacles for deploying EEG foundation models in BCI. A lightweight, backbone-agnostic adaptation layer that can be dropped onto existing models is practically useful if the gains hold under a clean online protocol. Strengths include breadth (six datasets, four backbones, five seeds, Wilcoxon tests), component ablations (Table 5), and promised code/data scripts. The contribution is primarily empirical and engineering-oriented rather than theoretical; its lasting value depends on whether the reported online improvements survive a stricter streaming evaluation and stronger external baselines.

major comments (3)
  1. [§3.4, Algorithm 1, §4.1] The central claim of superior online performance (Abstract; Tables 2–4; Fig. 1C) rests on an incompletely specified evaluation protocol. Algorithm 1 separates an “Online Training Phase” that updates with L_cls using true yt from an “Online Inference Phase” that predicts with fixed parameters, yet §4.1 states that the test set is streamed to evaluate continual adaptation. The manuscript never states whether the metric for sample t is computed before the gradient step that uses yt, after it, or on a held-out future window. If evaluation reuses the same labeled batch that drove the update, reported Bal.Acc./Kappa gains partly measure label-driven self-fitting rather than generalization to future stream samples. This ambiguity is load-bearing for the large jumps on SEED-V and BCIC-2A and must be resolved with an explicit predict-then-update (or delayed-label) protocol and corresponding re-ru
  2. [§4.1 Baselines; Tables 2–4] The only online baseline is NeuroOnline.a—the authors’ own method with MCL and CRM removed (§4.1, Tables 2–4). No comparison is made to established test-time or online adaptation methods (e.g., entropy minimization / TENT-style, BN adaptation, or simple online SGD with replay) that are standard in the CV/NLP literature the paper itself cites. Without such baselines it is impossible to tell whether the gains come from the proposed mechanisms or from any form of supervised online fine-tuning. At least two external online/TTA baselines should be added under the same causal streaming protocol.
  3. [§3.4 Eq. (4); Algorithm 1; §5 Limitations] The training loop (Eq. 4, Algorithm 1) assumes that the true label yt is immediately available for every online sample so that L_cls can be computed. The paper never evaluates the setting in which labels are delayed, sparse, or absent—common in real BCI deployments. Under pure unsupervised consistency the method reduces to L_cons alone, which is not reported. Either a delayed/semi-supervised experiment or a clear limitation statement quantifying how much of the gain depends on immediate labels is required for the “online settings” claim to be credible.
minor comments (5)
  1. [Fig. 1C; Abstract] Fig. 1C and the abstract claim average relative improvements (e.g., +11.52%, +17.85%) without stating which metric and which baseline set are averaged; a short caption or appendix table would avoid ambiguity.
  2. [Fig. 2; Eq. (2)] Notation for the consistency loss appears with garbled subscripts in Fig. 2 (L_ccc) while the text uses L_cons; unify.
  3. [Appendix F; Table 8] Hyperparameter tables (Tables 7–8) list online epochs = 3 and batch size 64; clarify how a batch is formed under a strictly causal stream (sliding window? single-sample updates?).
  4. [§2.2] Related Work §2.2 asserts that EEG non-stationarity is “more severe” than image/text without a quantitative citation or experiment; soften or support.
  5. [§5; NeurIPS Checklist] Checklist item 10 claims discussion of positive societal impacts; the main text §5 only briefly mentions future work. A short Broader Impacts paragraph would match the checklist claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: purely empirical online-adaptation framework whose reported gains are measured on held-out streaming subjects against independent baselines, with no derivation that reduces a claim to its own fitted inputs or self-citation.

full rationale

NeuroOnline is an engineering/empirical paper. Its central claims are performance numbers (Bal.Acc., Kappa, etc.) obtained by streaming held-out test subjects through the online loop of Algorithm 1 and comparing against offline fine-tuning and a stripped online baseline (NeuroOnline.a). The two mechanisms (multi-view consistency L_cons and context-aware modulation via prompt P) are defined by ordinary losses and a cross-attention modulator; none of the equations equate a claimed prediction to a quantity that was fitted from the same data. Self-citations are limited to using publicly released foundation-model backbones (CBraMod, LaBraM, EEGPT, BIOT) whose authors do not overlap with the present paper; those backbones supply features, not uniqueness theorems or load-bearing premises that force the online results. The evaluation protocol may be open to criticism about label timing, but that is an experimental-validity concern, not a circular reduction of a derivation to its inputs. Consequently the derivation chain is self-contained against external benchmarks and scores 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 2 invented entities

The central claim rests on standard EEG non-stationarity lore, the unproven semantic-preservation property of the chosen masks, the expressive power of a single learnable prompt, and the availability of online labels. Free parameters are ordinary training hyper-parameters; the invented entities are the two modules themselves, which have no independent external validation beyond the paper’s own tables.

free parameters (3)
  • λ (consistency-loss weight) = 0.25–1.0
    Balances L_cls against L_cons; sensitivity table shows performance peaks between 0.25–1.0 and is chosen per backbone/dataset via validation.
  • K (number of augmented views) = 2
    Fixed to 2 without extensive ablation; directly controls the multi-view consistency term.
  • online epochs / learning rate / batch size = 3 epochs, 1e-4, 64
    Online phase uses only 3 epochs, lr=1e-4, batch=64 (Table 8); these control how aggressively the model tracks the stream and are set by hand following offline practice.
assumptions (4)
  • domain assumption EEG signals are non-stationary; their statistics evolve over time and across subjects, producing distribution shifts that static models cannot handle.
    Stated as background in Introduction and Related Work; taken from the BCI literature without new measurement.
  • ad hoc to paper Temporal masking and frequency masking produce views that preserve task-relevant semantic content while supplying complementary perturbations.
    Section 3.2 asserts this property to justify the consistency loss; no formal or empirical verification of semantic preservation is supplied.
  • ad hoc to paper A single learnable context prompt interacting via cross-attention is sufficient to capture latent subject/task/environment context for representation modulation.
    Core design choice of CRM (Section 3.3); success is measured only by downstream accuracy inside the paper.
  • domain assumption Ground-truth labels yt arrive with each online sample and can be used for supervised updates.
    Required by Eq. 4 and Algorithm 1; never relaxed or tested under delayed/unsupervised feedback.
invented entities (2)
  • NeuroOnline (MCL + CRM joint framework)
    purpose: Unify pretraining–task alignment and continuous distribution adaptation for EEG foundation models.
    The combination is new; independent evidence is limited to the paper’s own online benchmarks.
  • Context prompt P + cross-attention modulator producing α, β
    purpose: Dynamically scale and shift backbone features according to evolving online context.
    Specific architectural module introduced in Section 3.3; no external falsifiable prediction or independent replication is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of NeuroOnline: Bridging Pretraining and Online Adaptation for EEG Foundation Models." pith.science (2026). https://pith.science/paper/OGETJPTN

@misc{pith2026260703925,
  author       = {Pith},
  title        = {Pith review of: NeuroOnline: Bridging Pretraining and Online Adaptation for EEG Foundation Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OGETJPTN}},
  note         = {Machine review of arXiv:2607.03925}
}
read the original abstract

EEG foundation models have shown strong potential in learning generalized representations across subjects and tasks. However, most existing approaches follow a pretraining-static deployment paradigm, which suffers from two key limitations: (1) misalignment between pretraining objectives and downstream tasks, and (2) limited adaptability to distribution shifts in online settings. We propose Online Neural Adaptation (NeuroOnline), a unified framework that enables continuous adaptation in online scenarios. NeuroOnline integrates two complementary mechanisms: (1) multi-view consistency learning, which enforces cross-view alignment to promote consistent and task-relevant representations, and (2) context-aware representation modulation, which leverages a learnable context prompt with cross-attention to dynamically adapt representations to evolving data distributions. Together, these mechanisms unify representation alignment and dynamic adaptation. Experiments on multiple EEG benchmarks show that NeuroOnline consistently outperforms strong baselines in online settings, achieving better performance under distribution shifts. Ablation and sensitivity studies further validate the necessity of each component and the effectiveness of the overall design.

Figures

Figures reproduced from arXiv: 2607.03925 by the authors.

Figure 1
Figure 1. Motivation. (A) Challenges: pretraining–task misalignment and limited adaptability under distribution shifts in online EEG settings. (B) Our paradigm (NeuroOnline): continual adaptation of foundation models to evolving neural dynamics. (C) Performance comparison: NeuroOnline consistently achieves relative improvements over the strongest offline baseline (off-FT-MLP) across multiple datasets and EEG foundation models… view at source ↗
Figure 2
Figure 2. Framework of NeuroOnline. (A) Overall pipeline: given streaming EEG data, Neu￾roOnline constructs multi-view augmentations, encodes representations with a backbone model, and integrates multi-view consistency learning (MCL) and context-aware representation modulation (CRM) for continuous online adaptation. (B) Multi-view consistency learning: representations from augmented views are aligned to mitigate pretraining–t… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 8 canonical work pages

  1. [1]

    A comprehensive review of eeg-based brain–computer interface paradigms.Journal of neural engineering, 16(1): 011001, 2019

    Reza Abiri, Soheil Borhani, Eric W Sellers, Yang Jiang, and Xiaopeng Zhao. A comprehensive review of eeg-based brain–computer interface paradigms.Journal of neural engineering, 16(1): 011001, 2019

    2019

  2. [2]

    Review of machine learning techniques for eeg based brain computer interface.Archives of Computational Methods in Engineering, 29(5):3001–3020, 2022

    Swati Aggarwal and Nupur Chugh. Review of machine learning techniques for eeg based brain computer interface.Archives of Computational Methods in Engineering, 29(5):3001–3020, 2022

    2022

  3. [3]

    A review of classification algorithms for eeg-based brain– computer interfaces: a 10 year update.Journal of neural engineering, 15(3):031005, 2018

    Fabien Lotte, Laurent Bougrain, Andrzej Cichocki, Maureen Clerc, Marco Congedo, Alain Rakotomamonjy, and Florian Yger. A review of classification algorithms for eeg-based brain– computer interfaces: a 10 year update.Journal of neural engineering, 15(3):031005, 2018

    2018

  4. [4]

    Past, present, and future of eeg-based bci applications.Sensors, 22(9):3331, 2022

    Kaido Värbu, Naveed Muhammad, and Yar Muhammad. Past, present, and future of eeg-based bci applications.Sensors, 22(9):3331, 2022

    2022

  5. [5]

    Recent advances in natural language processing via large pre-trained language models: A survey.ACM Comput

    Bonan Min, Hayley Ross, Elior Sulem, Amir Pouran Ben Veyseh, Thien Huu Nguyen, Oscar Sainz, Eneko Agirre, Ilana Heintz, and Dan Roth. Recent advances in natural language processing via large pre-trained language models: A survey.ACM Comput. Surv., 56(2), September 2023. ISSN 0360-0300. doi: 10.1145/3605943. URL https://doi.org/10. 1145/3605943

    doi:10.1145/3605943 2023

  6. [6]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners. In2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15979–15988, 2022. doi: 10.1109/CVPR52688.2022. 01553

    doi:10.1109/cvpr52688.2022 2022

  7. [7]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. InInternational conference on machine learning, pages 1597–1607. PmLR, 2020

    2020

  8. [8]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. InProceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers), pages 4171–4186, 2019

    2019

Show all 42 references
  1. [9]

    Eeg foundation models: a critical review of current progress and future directions.Journal of neural engineering, 23(2):021001, 2026

    Gayal Kuruppu, Neeraj Wagh, Vaclav Kremen, and Yogatheesan Varatharajah. Eeg foundation models: a critical review of current progress and future directions.Journal of neural engineering, 23(2):021001, 2026

    2026

  2. [10]

    A simple review of eeg foundation models: Datasets, advancements and future perspectives.arXiv preprint arXiv:2504.20069, 2025

    Junhong Lai, Jiyu Wei, Lin Yao, and Yueming Wang. A simple review of eeg foundation models: Datasets, advancements and future perspectives.arXiv preprint arXiv:2504.20069, 2025

    2025

  3. [11]

    Adabrain-bench: Benchmarking brain foundation models for brain-computer interface applications.arXiv preprint arXiv:2507.09882, 2025

    Jiamin Wu, Zichen Ren, Junyu Wang, Pengyu Zhu, Yonghao Song, Mianxin Liu, Qihao Zheng, Lei Bai, Wanli Ouyang, and Chunfeng Song. Adabrain-bench: Benchmarking brain foundation models for brain-computer interface applications.arXiv preprint arXiv:2507.09882, 2025

    2025 arXiv

  4. [12]

    Eeg-based emotion recognition: A state-of-the-art review of current trends and opportunities.Computational Intelligence and Neuroscience, 2020(1):8875426, 2020

    Nazmi Sofian Suhaimi, James Mountstephens, and Jason Teo. Eeg-based emotion recognition: A state-of-the-art review of current trends and opportunities.Computational Intelligence and Neuroscience, 2020(1):8875426, 2020. doi: https://doi.org/10.1155/2020/8875426. URL https://onl...

    2020 doi

  5. [13]

    A review of feature extraction and performance evaluation in epileptic seizure detection using eeg.Biomedical Signal Processing and Control, 57:101702, 2020

    Poomipat Boonyakitanont, Apiwat Lek-uthai, Krisnachai Chomtho, and Jitkomut Songsiri. A review of feature extraction and performance evaluation in epileptic seizure detection using eeg.Biomedical Signal Processing and Control, 57:101702, 2020. ISSN 1746-8094. doi: https: //doi...

    2020 doi

  6. [14]

    Al-Zoubi, Hossam Faris, and Ibrahim Aljarah

    Lakhan Dev Sharma, Vijay Kumar Bohat, Maria Habib, Ala’ M. Al-Zoubi, Hossam Faris, and Ibrahim Aljarah. Evolutionary inspired approach for mental stress detection using eeg signal.Expert Systems with Applications, 197:116634, 2022. ISSN 0957-4174. doi: https: //doi.org/10.1016...

    2022 doi

  7. [15]

    Shamim Hossain

    Syed Umar Amin, Mansour Alsulaiman, Ghulam Muhammad, Mohamed Amine Mekhtiche, and M. Shamim Hossain. Deep learning for eeg motor imagery classification based on multi-layer cnns feature fusion.Future Generation Computer Systems, 101:542–554, 2019. ISSN 0167- 739X. doi: https:/...

    2019 doi

  8. [16]

    Chrononet: A deep recurrent neural net- work for abnormal eeg identification

    Subhrajit Roy, Isabell Kiral-Kornek, and Stefan Harrer. Chrononet: A deep recurrent neural net- work for abnormal eeg identification. InArtificial Intelligence in Medicine: 17th Conference on Artificial Intelligence in Medicine, AIME 2019, Poznan, Poland, June 26–29, 2019, Pro...

    2019 doi

  9. [17]

    Wouter Biesmans, Neetha Das, Tom Francart, and Alexander Bertrand. Auditory-inspired speech envelope extraction methods for improved eeg-based auditory attention detection in a cocktail party scenario.IEEE Transactions on Neural Systems and Rehabilitation Engineering, 25(5):40...

    2017 doi

  10. [18]

    Aboalayon, Miad Faezipour, Wafaa S

    Khald Ali I. Aboalayon, Miad Faezipour, Wafaa S. Almuhammadi, and Saeid Moslehpour. Sleep stage classification using eeg signal analysis: A comprehensive survey and new investigation. Entropy, 18(9), 2016. ISSN 1099-4300. doi: 10.3390/e18090272. URL https://www.mdpi. com/1099-...

    2016 doi

  11. [19]

    Self-supervised learning: Generative or contrastive.IEEE Transactions on Knowledge and Data Engineering, 35(1):857–876, 2023

    Xiao Liu, Fanjin Zhang, Zhenyu Hou, Li Mian, Zhaoyu Wang, Jing Zhang, and Jie Tang. Self-supervised learning: Generative or contrastive.IEEE Transactions on Knowledge and Data Engineering, 35(1):857–876, 2023. doi: 10.1109/TKDE.2021.3090866

    2023 doi

  12. [20]

    Cbramod: A criss-cross brain foundation model for eeg decoding.arXiv preprint arXiv:2412.07236, 2024

    Jiquan Wang, Sha Zhao, Zhiling Luo, Yangxuan Zhou, Haiteng Jiang, Shijian Li, Tao Li, and Gang Pan. Cbramod: A criss-cross brain foundation model for eeg decoding.arXiv preprint arXiv:2412.07236, 2024

    2024

  13. [21]

    Large brain model for learning generic representations with tremendous eeg data in bci

    Wei-Bang Jiang, Liming Zhao, and Bao-liang Lu. Large brain model for learning generic representations with tremendous eeg data in bci. In B. Kim, Y . Yue, S. Chaudhuri, K. Fragkiadaki, M. Khan, and Y . Sun, editors,International Conference on Learning Representations, volume 2...

    2024

  14. [22]

    Non- stationarity in brain-computer interfaces: An analytical perspective.arXiv preprint arXiv:2512.15941, 2025

    Hubert Cecotti, Rashmi Mrugank Shah, Raksha Jagadish, and Toshihisa Tanaka. Non- stationarity in brain-computer interfaces: An analytical perspective.arXiv preprint arXiv:2512.15941, 2025

    2025

  15. [23]

    Challenge for affective brain-computer interfaces: non- stationary spatio-spectral eeg oscillations of emotional responses.Frontiers in human neuro- science, 13:366, 2019

    Yi-Wei Shen and Yuan-Pin Lin. Challenge for affective brain-computer interfaces: non- stationary spatio-spectral eeg oscillations of emotional responses.Frontiers in human neuro- science, 13:366, 2019

    2019

  16. [24]

    Haider Raza, Dheeraj Rathee, Shang-Ming Zhou, Hubert Cecotti, and Girijesh Prasad. Covariate shift estimation based adaptive ensemble learning for handling non-stationarity in motor imagery related eeg-based brain-computer interface.Neurocomputing, 343:154– 166, 2019. ISSN 092...

    2019 doi

  17. [25]

    Eegpt: Pretrained transformer for universal and reliable representation of eeg signals

    Guagnyu Wang, Wenchao Liu, Yuhong He, Cong Xu, Lin Ma, and Haifeng Li. Eegpt: Pretrained transformer for universal and reliable representation of eeg signals. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, editors,Advances in Neural Infor...

    2024 doi

  18. [26]

    Biot: Biosignal transformer for cross-data learning in the wild

    Chaoqi Yang, M Westover, and Jimeng Sun. Biot: Biosignal transformer for cross-data learning in the wild. In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors, Advances in Neural Information Processing Systems, volume 36, pages 78240–78260. Curran As...

    2023

  19. [27]

    Hoi, Doyen Sahoo, Jing Lu, and Peilin Zhao

    Steven C.H. Hoi, Doyen Sahoo, Jing Lu, and Peilin Zhao. Online learning: A comprehensive survey.Neurocomputing, 459:249–289, 2021. ISSN 0925-2312. doi: https://doi.org/10.1016/j. neucom.2021.04.112. URL https://www.sciencedirect.com/science/article/pii/ S0925231221006706

    2021 doi

  20. [28]

    Contrastive test-time adaptation

    Dian Chen, Dequan Wang, Trevor Darrell, and Sayna Ebrahimi. Contrastive test-time adaptation. In2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 295–305, 2022. doi: 10.1109/CVPR52688.2022.00039

    2022 doi

  21. [29]

    Online video understanding: Ovbench and videochat- online

    Zhenpeng Huang, Xinhao Li, Jiaqi Li, Jing Wang, Xiangyu Zeng, Cheng Liang, Tao Wu, Xi Chen, Liang Li, and Limin Wang. Online video understanding: Ovbench and videochat- online. InProceedings of the Computer Vision and Pattern Recognition Conference, pages 3328–3338, 2025

    2025

  22. [30]

    Continual learning of large language models: A comprehensive survey.ACM Comput

    Haizhou Shi, Zihao Xu, Hengyi Wang, Weiyi Qin, Wenyuan Wang, Yibin Wang, Zifeng Wang, Sayna Ebrahimi, and Hao Wang. Continual learning of large language models: A comprehensive survey.ACM Comput. Surv., 58(5), November 2025. ISSN 0360-0300. doi: 10.1145/3735633. URLhttps://doi...

    2025 doi

  23. [31]

    Tent: Fully test-time adaptation by entropy minimization.arXiv preprint arXiv:2006.10726, 2020

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization.arXiv preprint arXiv:2006.10726, 2020

    2006 arXiv

  24. [32]

    Mutual information-driven subject- invariant and class-relevant deep representation learning in bci.IEEE Transactions on Neural Networks and Learning Systems, 34(2):739–749, 2023

    Eunjin Jeon, Wonjun Ko, Jee Seok Yoon, and Heung-Il Suk. Mutual information-driven subject- invariant and class-relevant deep representation learning in bci.IEEE Transactions on Neural Networks and Learning Systems, 34(2):739–749, 2023. doi: 10.1109/TNNLS.2021.3100583

    2023 doi

  25. [33]

    Eeg temporal–spatial transformer for person identification.Scientific Reports, 12(1):14378, 2022

    Yang Du, Yongling Xu, Xiaoan Wang, Li Liu, and Pengcheng Ma. Eeg temporal–spatial transformer for person identification.Scientific Reports, 12(1):14378, 2022

    2022

  26. [34]

    Bci competition 2008–graz data set a.Institute for knowledge discovery (laboratory of brain- computer interfaces), Graz University of Technology, 16(1-6):1, 2008

    Clemens Brunner, Robert Leeb, Gernot Müller-Putz, Alois Schlögl, and Gert Pfurtscheller. Bci competition 2008–graz data set a.Institute for knowledge discovery (laboratory of brain- computer interfaces), Graz University of Technology, 16(1-6):1, 2008

    2008

  27. [35]

    David Steyrl, Reinhold Scherer, Josef Faller, and Gernot R Müller-Putz. Random forests in non-invasive sensorimotor rhythm brain-computer interfaces: a practical and convenient non-linear classifier.Biomedical Engineering/Biomedizinische Technik, 61(1):77–86, 2016

    2016

  28. [36]

    A large eeg dataset for studying cross-session variability in motor imagery brain-computer interface

    Jun Ma, Banghua Yang, Wenzheng Qiu, Yunzhe Li, Shouwei Gao, and Xinxing Xia. A large eeg dataset for studying cross-session variability in motor imagery brain-computer interface. Scientific Data, 9(1):531, 2022

    2022

  29. [37]

    Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals.circulation, 101(23):e215–e220, 2000

    Ary L Goldberger, Luis AN Amaral, Leon Glass, Jeffrey M Hausdorff, Plamen Ch Ivanov, Roger G Mark, Joseph E Mietus, George B Moody, Chung-Kang Peng, and H Eugene Stanley. Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic si...

    2000

  30. [38]

    Bci2000: a general-purpose brain-computer interface (bci) system.IEEE Transactions on biomedical engineering, 51(6):1034–1043, 2004

    Gerwin Schalk, Dennis J McFarland, Thilo Hinterberger, Niels Birbaumer, and Jonathan R Wolpaw. Bci2000: a general-purpose brain-computer interface (bci) system.IEEE Transactions on biomedical engineering, 51(6):1034–1043, 2004

    2004

  31. [39]

    A large finer-grained affective computing eeg dataset.Scientific Data, 10(1):740, 2023

    Jingjing Chen, Xiaobin Wang, Chen Huang, Xin Hu, Xinke Shen, and Dan Zhang. A large finer-grained affective computing eeg dataset.Scientific Data, 10(1):740, 2023

    2023

  32. [40]

    Wei Liu, Jie-Lin Qiu, Wei-Long Zheng, and Bao-Liang Lu. Comparing recognition performance and robustness of multimodal deep learning models for multimodal emotion recognition.IEEE Transactions on Cognitive and Developmental Systems, 14(2):715–729, 2021

    2021

  33. [41]

    Limitations

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.Iclr, 1(2):3, 2022. 12 Algorithm 1The Pipeline of NeuroOnline (1) Online Training Phase: Input:Observed sample...

    2022

  34. [42]

    Guidelines: • The answer [N/A] means that the paper does not involve crowdsourcing nor research with human subjects

    Institutional review board (IRB) approvals or equivalent for research with human subjects Question: Does the paper describe potential risks incurred by study participants, whether such risks were disclosed to the subjects, and whether Institutional Review Board (IRB) approvals...

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.