Backpropagation-Free Test-Time Adaptation for Lightweight EEG-Based Brain-Computer Interfaces

Dongrui Wu; Feng Wan; Jiayi Ouyang; Siyang Li; Tianwang Jia; Zhenyao Cui; Ziwei Wang

REVIEW 3 major objections 5 minor 45 references

Backpropagation-Free Test-Time Adaptation for Lightweight EEG-Based Brain-Computer Interfaces

T0 review · 3 major / 5 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read This paper claims that test-time adaptation of EEG decoding models can be done without backpropagation by aggregating weighted predictions over label-preserving transformations, matching or beating gradient-based adaptation on five datasets

desk verdict A practical, backprop-free TTA for EEG that mostly delivers; theory oversold and target-domain ranker calibration under-evidenced. read the letter →

arxiv 2601.07556 v2 pith:XVGMW53L submitted 2026-01-12 cs.HC cs.AI

Siyang Li , Jiayi Ouyang , Zhenyao Cui , Ziwei Wang , Tianwang Jia , Feng Wan , Dongrui Wu This is my paper · ORCID

classification cs.HCcs.AI

keywords test-timeadaptationEEGdecodingbrain-computerinterfacebackpropagation-freelearning-to-rankvariancereductionmotorimagerydriverdrowsiness

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

BFT (Backpropagation-Free Transformations) applies several label-preserving transformations to each test trial, runs them through a frozen pretrained EEG decoder, and combines the resulting predictions with reliability weights learned by a ranking module. This provides calibration-free and backpropagation-free test-time adaptation for both motor-imagery classification and driver-drowsiness regression. The paper proves that this weighted aggregation reduces prediction variance, and experiments on five EEG datasets show it matches or exceeds the accuracy of backpropagation-based T-TIME while being robust to test-time noise and 8-bit quantization. The practical payoff is a lightweight, privacy-preserving, task-agnostic adaptation method suitable for plug-and-play BCIs on resource-constrained devices.

What carries the argument

The central mechanism is the combination of (1) label-preserving test-time transformations—noise addition, amplitude scaling, frequency shift, sliding windows, and monotone-dropout feature masks—that produce multiple stochastic branches per trial, and (2) a learning-to-rank module r(·), trained with a mapping module m(·) that projects task losses into a rank-like space, which assigns a reliability weight to each branch. The weighted aggregation (convex combination with temperature sharpening for classification, top-half averaging for regression) is what carries the variance-reduction argument, and the ranking module is what differentiates BFT from naive averaging such as Aug-Mean or MC Dropo

What would settle it

A concrete test: on a held-out target subject, intentionally choose one transformation that is label-destroying (for example, a frequency shift that moves the signal away from the subject's discriminative band). If BFT's accuracy then drops below the unweighted Aug-Mean baseline or below the best single transformation, the ranking module is amplifying an unreliable branch and the variance-reduction guarantee does not hold in that regime. Alternatively, empirically estimate Var(ŵf) and V0 on target trials with known labels and check whether the inequality Var(ŵf) < V0 holds; if it is violated f

Watch

Extended reading notes

Core claim

The paper's central claim is that prediction variance at test time can be suppressed without any model update: generate multiple predictions for a single test sample by applying structured transformations (knowledge-guided augmentations in BFT-A, or deterministic dropout-style feature masks in BFT-D), then combine them with weights from a learned ranking module. The authors prove in Theorem 2 that the aggregated variance Var(ŵf(x)) is smaller than the single-shot variance V0 whenever the effective number of branches satisfies Keff > κ(1−ρmax)/(1−κρmax), where κ bounds the worst-branch variance relative to V0 and ρmax bounds the absolute inter-branch correlation. Empirically, BFT-A reaches 79

Load-bearing premise

The load-bearing premise is that the transformations applied at test time preserve the task-relevant label for the target subject, and that the reliability ranking learned from source-subject task losses continues to assign trustworthy weights on the shifted target domain.

Editorial extensions

If this is right

Plug-and-play EEG BCIs can adapt to a new user's signal stream without per-use calibration, backpropagation, or access to the model's internal parameters.
The same forward-only aggregation works for regression (driver drowsiness estimation) and classification, covering tasks that most TTA methods exclude.
BFT remains effective after post-training 8-bit quantization, making it compatible with edge-device and CPU-only deployments where backpropagation-based TTA is infeasible.
The theoretical variance bound provides a formal, parameter-free justification for why test-time ensembling of transformations reduces prediction uncertainty, independent of the specific backbone architecture.
The method's robustness to temporal test-time noise, verified on all five datasets, suggests it can tolerate realistic EEG artifacts without special handling.

Reading between the lines

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

The variance-reduction theorem is not EEG-specific: any frozen black-box model with label-preserving perturbations and a reasonable reliability ranking could benefit from this aggregation strategy, so the method likely transfers to other biosignal or non-biosignal domains.
The ranking module is trained on source-domain task losses only; on a target domain where transformations are not label-preserving or the ranker misorders branches, weighted aggregation could amplify an unreliable branch—a testable extension is to calibrate the ranker on a small labeled target-sample set.
BFT-D's deterministic feature masks provide a parameter-free alternative that requires no domain-specific augmentation design, suggesting the method can be dropped into any pretrained network that has a dropout layer.
Because the paper evaluates only a single backbone (EEGNet), a natural extension is to test BFT with transformer-based or hybrid EEG decoders, where the label-preservation assumption on transformations may behave differently.

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.

Desk Editor's Note

read the letter

Short version: this is a solid applied paper. The BFT idea—aggregate over test-time transformations with weights from a source-trained ranking module, no backprop—is genuinely new in the EEG TTA space, and they show it works on five datasets, across classification and regression, and even under quantization. That's a real contribution for anyone deploying BCIs on edge hardware.

What I like: the method is clearly described, the ablations isolating the mapping module are honest, and the comparison set is broad, including backprop-based TTA and plain Aug-Mean. The robustness experiments with temporal and spatial noise are a nice touch. The fact that BFT-A and BFT-D beat uniform averaging on most datasets gives some empirical support for the ranking module, which is the core novelty.

Soft spots: the theory in Section IV is just a standard variance decomposition. Theorem 2 shows that averaging reduces variance under assumptions on branch variances and correlations, but it doesn't say anything about the learned weights being better than uniform—it bounds variance for any weights. So the theoretical justification is overclaiming, but it isn't load-bearing; the empirical results carry the paper.

The bigger gap is the one the stress-test flags: the ranking module is trained on source-domain task losses, and the paper never shows it stays calibrated on the target subject. There's one NDCG figure for one subject, and the rest is aggregate accuracy. If the ranker misorders transformations on a specific subject, weighted aggregation could underperform uniform averaging. The aggregate results suggest this isn't happening on average, but per-subject variance is hidden. I'd want to see per-subject gains over Aug-Mean and ideally a calibration check on target branches. This is a fixable paper-level weakness, not a fatal one.

Also: no significance tests across the three seeds, and some of the reported differences (e.g., BFT-A 79.03 vs MEMO 78.19) may be within noise. A paired test would help.

Overall: this paper deserves a serious referee. It's an incremental but useful advance, clearly explained, with reproducible code. It should be sent to review, with the main revision being to temper the theory and shore up the target-domain evaluation.

Referee Report

3 major / 5 minor

Summary. The paper proposes BFT, a backpropagation-free test-time adaptation method for EEG-based BCIs. At inference, BFT applies multiple transformations to each test trial (input augmentations in BFT-A, or deterministic feature-dropout masks in BFT-D), obtains several forward-pass predictions, and aggregates them using reliability weights produced by a learning-to-rank module. The ranking module is trained on source-domain task losses, with an auxiliary mapping module that converts soft scores into rank-like values. A variance-decomposition argument (Theorems 1 and 2) is given as theoretical justification. Experiments on three motor-imagery classification datasets and two driver-drowsiness regression datasets under leave-one-subject-out evaluation report accuracy/CC/RMSE improvements over backpropagation-free baselines, competitiveness with backpropagation-based TTA, robustness to simulated test-time noise, and compatibility with 8-bit quantization.

Significance. If the empirical claims hold, BFT is a useful practical contribution: it is black-box, forward-only, task-agnostic, and works for both classification and regression, addressing a real deployment gap for lightweight EEG BCIs. The evaluation is credible in important respects: held-out subjects, public benchmark datasets, multiple random seeds, released code, and direct CPU/GPU latency measurements. The variance decomposition is mathematically correct but standard and does not specifically justify the learned weighting; the novelty is mainly empirical. The paper's central claim that BFT is 'much more practical' than current TTA approaches is plausible but needs stronger evidence that the learned ranking, rather than simple averaging, transfers to target subjects.

major comments (3)

[§IV-D, Theorem 2 (Eqs. 19–21)] The theorem bounds Var(ˆf_w) for arbitrary probability weights and shows that any non-degenerate averaging reduces variance relative to single-shot inference under Assumptions 1–2. It does not show that the learned reliability weights from r(·) are better than uniform weights, nor that reducing this variance improves classification/regression risk. The statement in §IV-D that the learning-to-rank module 'suppresses unreliable branches' and 'helps uncertainty reduction' is therefore not a logical consequence of the theorem. Please add a target-domain analysis comparing learned weighted aggregation to uniform Aug-Mean/MC Dropout in terms of per-subject risk and effective Keff; if the difference is not significant, temper the claim.
[§V-G, Fig. 8(b)] The only target-domain evaluation of the ranking module is one subject (Driving S1) with NDCG@6 = 0.611, and the classification panel in Fig. 8(a) is also a single subject. Since the contribution over naive transformation ensembles is precisely the learned reliability weighting, per-subject calibration of r(·) on the target is load-bearing. Aggregate accuracies in Tables II–IV can mask subjects where the ranker misorders transformations. Please report per-subject NDCG or rank correlation for all subjects/datasets, and a statistical comparison (e.g., paired test across seeds) of BFT against uniform Aug-Mean and MC Dropout. Without this, the claim that the ranking module 'enhances weighting' is not established.
[§III-B / §IV-A, Definition 1] The theory assumes label-preserving test-time randomization, but this is not verified for the target domain. For BFT-A, transformations such as frequency shift and sliding windows are plausible for motor imagery, yet a subject whose discriminative rhythm lies in the shifted band could receive misleading branches. The variance decomposition does not capture the resulting bias error. Please either verify label preservation quantitatively on held-out target data (e.g., by comparing branch predictions to oracle labels during evaluation), or weaken the theoretical claim to prediction-variance reduction under the stated assumption.

minor comments (5)

[§V-C] The text says K=12 for BFT-A, but the enumeration (Identity, 3 Scale factors, Noise, Freq, 5 Slide windows) sums to 11. Please reconcile the count or clarify which transformation is missing.
[§IV-A, Eq. (9)] f is defined as a scalar E[y|x; ζ], but classification outputs are softmax vectors. State whether the variance is computed on a scalar summary (e.g., confidence) or per logit/probability, since this affects the interpretation of V0.
[§III-C, Eq. (6)] Clarify the domain and codomain of the mapping module m(·): the text says it maps softmax-normalized losses in [0,1] to rank-like values in [1,K], but Eq. (6) applies m to the weight vector w_i. Please make the input/output spaces explicit.
[§V-H, Table VI] T-TIME is reported as N/A under quantized CPU. State explicitly why it is not applicable (e.g., backpropagation through quantized weights) and whether this is part of the practicality argument.
[Figures 2–3] Some labels in Figures 2 and 3 appear garbled in the PDF. Ensure the vector graphics are readable in the final version.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: BFT's components are fit only on source data, test-time aggregation uses unseen target inputs, and the variance bound is a standard inequality with un-fitted assumptions.

full rationale

The derivation is self-contained. The ranking module r(·) is trained on labeled source data via Eq. (6), with ground-truth ranks derived from the task module's source losses; at inference it is evaluated on held-out target inputs through Eqs. (7)–(8), so the reliability weights are not constructed from the test-time quantity they improve. The protocol is leave-one-subject-out ('No information from the test set was accessible during the training phase'), and the comparison includes external baselines and unweighted ablations (Aug-Mean, MC Dropout), so the gain from learned weighting is not forced by construction. The variance analysis (Lemma 1, Theorems 1–2) is a standard decomposition; Theorem 2's sufficient condition Keff > κ(1−ρmax)/(1−κρmax) uses κ and ρmax as assumptions, not fitted constants, so it is conditional rather than definitional. The only same-author citation used for background/baselines is [9] (T-TIME); it is not load-bearing. The paper itself flags a limitation in Fig. 8(b): NDCG@6 = 0.611 on one Driving subject and 'the variation across trials was substantial', and Section IV asserts 'the conditions for uncertainty reduction are approximately satisfied in practice' without measuring κ/ρmax; these are empirical-validation concerns, not circularity.

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

The method's load-bearing assumptions are domain assumptions about label preservation and transferability of the learned ranker, plus unverified structural assumptions in the theoretical bound. No new physical entities are introduced.

free parameters (4)

Number of transformations K = K=12 (BFT-A), K=10 (BFT-D)
K is chosen by the authors and determines the number of branches; no sensitivity analysis is reported, and it also sets the dropout rate p=1/K in BFT-D.
Temperature τ = 0.5
Sharpening temperature in Eq. 7; set to 0.5 without ablation.
Transformation hyperparameters = e.g., scale factors [0.9,1.1,1.2], sliding windows, noise variance
The specific augmentation parameters are listed in Section V-C but no search protocol or ablation is given.
Synthetic data distributions for mapping module = Four choices in Section V-B
The mapping module m(·) is pretrained on synthetic samples drawn from four distributions; these choices affect the rank projection.

assumptions (4)

domain assumption Test-time transformations are label-preserving
Definition 1 assumes 'label-preserving test-time randomization'; if noise/scaling/frequency shifts change the true label, aggregating their predictions could bias the result. This is asserted, not validated.
domain assumption The ranking module trained on source-domain task losses transfers to target domain
Section III-C trains r(·) on source data; Section V evaluates on held-out subjects without checking calibration of reliability scores under shift.
domain assumption Reducing prediction variance improves accuracy/transferability
Section IV proves variance reduction of the aggregated prediction, but no theorem links lower variance to lower error under domain shift; this link is asserted.
ad hoc to paper Homogeneous/heterogeneous variance and correlation bounds hold
Assumptions 1–2 and the condition ρmax < 1/κ are introduced to derive Theorems 1–2; the authors assert they are 'approximately satisfied in practice' without measurement.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Backpropagation-Free Test-Time Adaptation for Lightweight EEG-Based Brain-Computer Interfaces." pith.science (2026). https://pith.science/paper/XVGMW53L

@misc{pith2026260107556,
  author       = {Pith},
  title        = {Pith review of: Backpropagation-Free Test-Time Adaptation for Lightweight EEG-Based Brain-Computer Interfaces},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XVGMW53L}},
  note         = {Machine review of arXiv:2601.07556}
}

read the original abstract

Electroencephalogram (EEG)-based brain-computer interfaces (BCIs) face significant deployment challenges due to inter-subject variability, signal non-stationarity, and computational constraints. While test-time adaptation (TTA) mitigates distribution shifts under online data streams without per-use calibration sessions, existing TTA approaches heavily rely on explicitly defined loss objectives that require backpropagation for updating model parameters, which incurs computational overhead, privacy risks, and sensitivity to noisy data streams. This paper proposes Backpropagation-Free Transformations (BFT), a TTA approach for EEG decoding that eliminates such issues. BFT applies multiple sample-wise transformations of knowledge-guided augmentations or approximate Bayesian inference to each test trial, generating multiple prediction scores for a single test sample. A learning-to-rank module enhances the weighting of these predictions, enabling robust aggregation for uncertainty suppression during inference under theoretical justifications. Extensive experiments on five EEG datasets of motor imagery classification and driver drowsiness regression tasks demonstrate the effectiveness, versatility, robustness, and efficiency of BFT. This research enables lightweight plug-and-play BCIs on resource-constrained devices, broadening the real-world deployment of decoding algorithms for EEG-based BCI.

Figures

Figures reproduced from arXiv: 2601.07556 by the authors.

**Figure 2.** Two types of transformations. (a) BFT-A; and (b) BFT- [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Training and inference of the ranking module, and pre [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (6 more)

**Figure 4.** Figure 4: D. Heterogeneous Variance Case In the heterogeneous-variance setting, the ensemble uncertainty is mainly affected by three factors: the worst-branch noise κ, the similarity between branches ρmax, and how spread the weights are Keff [PITH_FULL_IMAGE:figures/full_fig_p…

**Figure 4.** Figure 4: Illustration of uncertainty reduction achieved thr [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: Two types of test-time noise, using an EEG trial from Z [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

**Figure 6.** Figure 6: Accuracy (%) under temporal and spatial noise during [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]

**Figure 7.** Figure 7: CCs and RMSEs under temporal and spatial noise during [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]

**Figure 8.** Figure 8: Evaluation of the learning-to-rank module for aggre [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 2 linked inside Pith

[1]

Interface, interact ion, and intel- ligence in generalized brain–computer interfaces,

X. Gao, Y . Wang, X. Chen, and S. Gao, “Interface, interact ion, and intel- ligence in generalized brain–computer interfaces,” Trends in Cognitive Sciences, vol. 25, no. 8, pp. 671–684, 2021

2021
[2]

Transfer learning for motor imagery based brain-computer interfaces: A tutorial,

D. Wu, X. Jiang, and R. Peng, “Transfer learning for motor imagery based brain-computer interfaces: A tutorial,” Neural Networks, vol. 153, pp. 235–253, 2022

2022
[3]

Affective brain–comp uter interfaces (aBCIs): A tutorial,

D. Wu, B.-L. Lu, B. Hu, and Z. Zeng, “Affective brain–comp uter interfaces (aBCIs): A tutorial,” Proc. of the IEEE , vol. 111, no. 10, pp. 1314–1332, 2023

2023
[4]

A survey on dro wsiness detection: Modern applications and methods,

B. Fu, F. Boutros, C.-T. Lin, and N. Damer, “A survey on dro wsiness detection: Modern applications and methods,” IEEE Trans. Intelligent V ehicles, vol. 9, no. 11, pp. 7279–7300, 2024

2024
[5]

A fully on-line adaptive BCI,

C. Vidaurre, A. Schlogl, R. Cabeza, R. Scherer, and G. Pfu rtscheller, “A fully on-line adaptive BCI,” IEEE Trans. Biomedical Engineering , vol. 53, no. 6, pp. 1214–1219, 2006

2006
[6]

A comprehensive survey on transfer learning,

F. Zhuang, Z. Qi, K. Duan, D. Xi, Y . Zhu, H. Zhu, H. Xiong, an d Q. He, “A comprehensive survey on transfer learning,” Proc. of the IEEE , vol. 109, no. 1, pp. 43–76, 2021

2021
[7]

A comprehensive survey on tes t-time adaptation under distribution shifts,

J. Liang, R. He, and T. Tan, “A comprehensive survey on tes t-time adaptation under distribution shifts,” Int’l Journal of Computer Vision , vol. 133, pp. 31–64, 2025

2025
[8]

I n search of lost online test-time adaptation: A survey,

Z. Wang, Y . Luo, L. Zheng, Z. Chen, S. Wang, and Z. Huang, “I n search of lost online test-time adaptation: A survey,” Int’l Journal of Computer Vision, vol. 133, p. 1106–1139, 2024

2024

Show all 45 references

[9]

T-TIME: Test-ti me information maximization ensemble for plug-and-play BCIs ,

S. Li, Z. Wang, H. Luo, L. Ding, and D. Wu, “T-TIME: Test-ti me information maximization ensemble for plug-and-play BCIs ,” IEEE Trans. Biomedical Engineering , vol. 71, no. 2, pp. 423–432, 2024

2024
[10]

Learn ing under concept drift: A review,

J. Lu, A. Liu, F. Dong, F. Gu, J. Gama, and G. Zhang, “Learn ing under concept drift: A review,” IEEE Trans. Knowledge and Data Engineering , vol. 31, no. 12, pp. 2346–2363, 2019

2019
[11]

Beyond model adaptation at test time: A survey,

Z. Xiao and C. G. Snoek, “Beyond model adaptation at test time: A survey,” arXiv preprint arXiv:2411.03687 , 2024

2024 arXiv
[12]

Improving robustness against common corruptio ns by covariate shift adaptation,

S. Schneider, E. Rusak, L. Eck, O. Bringmann, W. Brendel , and M. Bethge, “Improving robustness against common corruptio ns by covariate shift adaptation,” in Proc. Advances in Neural Information Processing Systems, V ancouver, Canada, Dec. 2020, pp. 11 539–11 551

2020
[13]

Tent: Fully test-time adaptation by entropy minimization,

D. Wang, E. Shelhamer, S. Liu, B. Olshausen, and T. Darre ll, “Tent: Fully test-time adaptation by entropy minimization,” in Int’l Conf. Learning Representations, Vienna, Austria, May. 2021

2021
[14]

Transfer learning for brain-computer i nterfaces: A Euclidean space data alignment approach,

H. He and D. Wu, “Transfer learning for brain-computer i nterfaces: A Euclidean space data alignment approach,” IEEE Trans. Biomedical Engineering, vol. 67, no. 2, pp. 399–410, 2020

2020
[15]

Pseudo-Label: The simple and efﬁcient semi -supervised learning method for deep neural networks,

D.-H. Lee, “Pseudo-Label: The simple and efﬁcient semi -supervised learning method for deep neural networks,” in Proc. Int’l Conf. Machine Learning W orkshops, Atlanta, GA, Jun. 2013, pp. 1322–1333

2013
[16]

Towards stable test-time adaptation in dynamic wild world ,

S. Niu, J. Wu, Y . Zhang, Z. Wen, Y . Chen, P . Zhao, and M. Tan , “Towards stable test-time adaptation in dynamic wild world ,” in Int’l Conf. Learning Representations , Kigali, Rwanda, May. 2023

2023
[17]

MEMO: Test time robu stness via adaptation and augmentation,

M. M. Zhang, S. Levine, and C. Finn, “MEMO: Test time robu stness via adaptation and augmentation,” in Proc. Advances in Neural Information Processing Systems, New Orleans, LA, Nov. 2022, pp. 38 629–38 642

2022
[18]

Quantization and training of neural n etworks for efﬁcient integer-arithmetic-only inference,

B. Jacob, S. Kligys, B. Chen, M. Zhu, M. Tang, A. Howard, H . Adam, and D. Kalenichenko, “Quantization and training of neural n etworks for efﬁcient integer-arithmetic-only inference,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition, Salt Lake City, UT, Jun. 2018, ...

2018
[19]

Q- EEGNet: an energy-efﬁcient 8-bit quantized parallel EEGNe t implemen- tation for edge motor-imagery brain-machine interfaces,

T. Schneider, X. Wang, M. Hersche, L. Cavigelli, and L. B enini, “Q- EEGNet: an energy-efﬁcient 8-bit quantized parallel EEGNe t implemen- tation for edge motor-imagery brain-machine interfaces,” in Proc. IEEE Int’l Conf. Smart Computing , Bologna, Italy, Sep. 2020, pp. 284–289

2020
[20]

Privacy-preserving brain–com puter interfaces: A systematic review,

K. Xia, W. Duch, Y . Sun, K. Xu, W. Fang, H. Luo, Y . Zhang, D. Sang, X. Xu, F.-Y . Wang, and D. Wu, “Privacy-preserving brain–com puter interfaces: A systematic review,” IEEE Trans. Computational Social Systems, vol. 10, no. 5, pp. 2312–2324, 2023

2023
[21]

Exploring model inversion attacks in the black-box setting,

A. Dionysiou, V . V assiliades, and E. Athanasopoulos, “ Exploring model inversion attacks in the black-box setting,” in Proc. Privacy Enhancing Technologies, Washington, DC, Jul. 2023, pp. 190–206

2023
[22]

On pitfalls of test -time adaptation,

H. Zhao, Y . Liu, A. Alahi, and T. Lin, “On pitfalls of test -time adaptation,” in Proc. Int’l Conf. Machine Learning , Honolulu, HI, Jul. 2023

2023
[23]

A survey on negati ve transfer,

W. Zhang, L. Deng, L. Zhang, and D. Wu, “A survey on negati ve transfer,” IEEE/CAA Journal of Automatica Sinica , vol. 10, no. 2, pp. 305–329, 2023. 14

2023
[24]

Data augmentation for self-pa ced motor imagery classiﬁcation with C-LSTM,

D. Freer and G.-Z. Y ang, “Data augmentation for self-pa ced motor imagery classiﬁcation with C-LSTM,” Journal of Neural Engineering , vol. 17, no. 1, p. 016041, 2020

2020
[25]

Dropout as a Bayesian approxi mation: Representing model uncertainty in deep learning,

Y . Gal and Z. Ghahramani, “Dropout as a Bayesian approxi mation: Representing model uncertainty in deep learning,” in Proc. Int’l Conf. Machine Learning , New Y ork City, NY , Jun. 2016, pp. 1050–1059

2016
[26]

Learning to rank for information retrieval,

T.-Y . Liu et al., “Learning to rank for information retrieval,” F oundations and Trends in Information Retrieval , vol. 3, no. 3, pp. 225–331, 2009

2009
[27]

S oDeep: a sorting deep net to learn ranking loss surrogates,

M. Engilberge, L. Chevallier, P . P´ erez, and M. Cord, “S oDeep: a sorting deep net to learn ranking loss surrogates,” in Proc. IEEE/CVF Conf. Computer Vision and Pattern Recognition , Long Beach, CA, Jun. 2019, pp. 10 792–10 801

2019
[28]

Zhou, Ensemble Methods: F oundations and Algorithms

Z.-H. Zhou, Ensemble Methods: F oundations and Algorithms . CRC Press, 2012

2012
[29]

A product fuz zy convo- lutional network for detecting driving fatigue,

G. Du, S. Long, C. Li, Z. Wang, and P . X. Liu, “A product fuz zy convo- lutional network for detecting driving fatigue,” IEEE Trans. Cybernetics, vol. 53, no. 7, pp. 4175–4188, 2023

2023
[30]

EEG-based driver drowsiness es timation using feature weighted episodic training,

Y . Cui, Y . Xu, and D. Wu, “EEG-based driver drowsiness es timation using feature weighted episodic training,” IEEE Trans. Neural Systems and Rehabilitation Engineering , vol. 27, no. 11, pp. 2263–2273, 2019

2019
[31]

Research on vehicl e-based driver status/performance monitoring: Development, validation, and reﬁnement of algorithms for detection of driver drowsiness. ﬁnal repo rt,

W. W. Wierwille and L. A. Ellsworth, “Research on vehicl e-based driver status/performance monitoring: Development, validation, and reﬁnement of algorithms for detection of driver drowsiness. ﬁnal repo rt,” U.S. Department of Transportation, National Highway Trafﬁc Saf ety Adm...

1994
[32]

A fully automa ted trial selection method for optimization of motor imagery ba sed brain- computer interface,

B. Zhou, X. Wu, Z. Lv, L. Zhang, and X. Guo, “A fully automa ted trial selection method for optimization of motor imagery ba sed brain- computer interface,” PloS One , vol. 11, no. 9, p. e0162657, 2016

2016
[33]

Review of the BCI competition IV,

M. Tangermann, K.-R. M¨ uller, A. Aertsen, N. Birbaumer , C. Braun, C. Brunner, R. Leeb, C. Mehring, K. J. Miller, G. R. M¨ uller-P utz et al., “Review of the BCI competition IV,” Frontiers in Neuroscience, vol. 6, p. 55, 2012

2012
[34]

Deep learning with convolutional neural networks for EEG d ecoding and visualization,

R. T. Schirrmeister, J. T. Springenberg, L. D. J. Fieder er, M. Glasstetter, K. Eggensperger, M. Tangermann, F. Hutter, W. Burgard, and T . Ball, “Deep learning with convolutional neural networks for EEG d ecoding and visualization,” Human Brain Mapping , vol. 38, no. 11, pp. ...

2017
[35]

Kine sthesia in a sustained-attention driving task,

C.-H. Chuang, L.-W. Ko, T.-P . Jung, and C.-T. Lin, “Kine sthesia in a sustained-attention driving task,” Neuroimage, vol. 91, pp. 187–202, 2014

2014
[36]

A multimodal approach to esti mating vig- ilance using EEG and forehead EOG,

W.-L. Zheng and B.-L. Lu, “A multimodal approach to esti mating vig- ilance using EEG and forehead EOG,” Journal of Neural Engineering , vol. 14, no. 2, p. 026017, 2017

2017
[37]

EEGNet: A compact convolutional neural net work for EEG-based brain-computer interfaces,

V . J. Lawhern, A. J. Solon, N. R. Waytowich, S. M. Gordon, C. P . Hung, and B. J. Lance, “EEGNet: A compact convolutional neural net work for EEG-based brain-computer interfaces,” Journal of Neural Engineering , vol. 15, no. 5, p. 056013, 2018

2018
[38]

Long short-term mem ory,

S. Hochreiter and J. Schmidhuber, “Long short-term mem ory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997

1997
[39]

Channel reﬂecti on: Knowledge-driven data augmentation for EEG-based brain–c omputer interfaces,

Z. Wang, S. Li, J. Luo, J. Liu, and D. Wu, “Channel reﬂecti on: Knowledge-driven data augmentation for EEG-based brain–c omputer interfaces,” Neural Networks , vol. 176, p. 106351, 2024

2024
[40]

Time–frequency trans form based EEG data augmentation for brain–computer interfaces,

Z. Wang, S. Li, X. Chen, and D. Wu, “Time–frequency trans form based EEG data augmentation for brain–computer interfaces,” Knowledge- Based Systems , vol. 311, p. 113074, 2025

2025
[41]

DARE-GRAM: Unsupervis ed domain adaptation regression by aligning inverse GRAM matrices,

I. Nejjar, Q. Wang, and O. Fink, “DARE-GRAM: Unsupervis ed domain adaptation regression by aligning inverse GRAM matrices,” in Proc. IEEE/CVF Conf. Computer Vision and Pattern Recognition , V ancouver, Canada, Jun. 2023, pp. 11 744–11 754

2023
[42]

A comp rehensive survey on artiﬁcial intelligence empowered edge computing on consumer electronics,

J.-H. Syu, J. C.-W. Lin, G. Srivastava, and K. Y u, “A comp rehensive survey on artiﬁcial intelligence empowered edge computing on consumer electronics,” IEEE Trans. Consumer Electronics , vol. 69, no. 4, pp. 1023–1034, 2023

2023
[43]

Deep learning with edge computing: A review,

J. Chen and X. Ran, “Deep learning with edge computing: A review,” Proc. of the IEEE , vol. 107, no. 8, pp. 1655–1674, 2019

2019
[44]

An accurate EEGNet-based motor-imagery brain–computer i nterface for low-power edge computing,

X. Wang, M. Hersche, B. T ¨ omekce, B. Kaya, M. Magno, and L . Benini, “An accurate EEGNet-based motor-imagery brain–computer i nterface for low-power edge computing,” in IEEE Int’l Symposium on Medical Measurements and Applications , Bari, Italy, Jul. 2020, pp. 1–6

2020
[45]

CMSIS-NN: Efﬁcient neu ral network kernels for ARM Cortex-M CPUs,

L. Lai, N. Suda, and V . Chandra, “CMSIS-NN: Efﬁcient neu ral network kernels for ARM Cortex-M CPUs,” arXiv preprint arXiv:1801.06601 , 2018

2018 arXiv

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.

[1] [1]

Interface, interact ion, and intel- ligence in generalized brain–computer interfaces,

X. Gao, Y . Wang, X. Chen, and S. Gao, “Interface, interact ion, and intel- ligence in generalized brain–computer interfaces,” Trends in Cognitive Sciences, vol. 25, no. 8, pp. 671–684, 2021

2021

[2] [2]

Transfer learning for motor imagery based brain-computer interfaces: A tutorial,

D. Wu, X. Jiang, and R. Peng, “Transfer learning for motor imagery based brain-computer interfaces: A tutorial,” Neural Networks, vol. 153, pp. 235–253, 2022

2022

[3] [3]

Affective brain–comp uter interfaces (aBCIs): A tutorial,

D. Wu, B.-L. Lu, B. Hu, and Z. Zeng, “Affective brain–comp uter interfaces (aBCIs): A tutorial,” Proc. of the IEEE , vol. 111, no. 10, pp. 1314–1332, 2023

2023

[4] [4]

A survey on dro wsiness detection: Modern applications and methods,

B. Fu, F. Boutros, C.-T. Lin, and N. Damer, “A survey on dro wsiness detection: Modern applications and methods,” IEEE Trans. Intelligent V ehicles, vol. 9, no. 11, pp. 7279–7300, 2024

2024

[5] [5]

A fully on-line adaptive BCI,

C. Vidaurre, A. Schlogl, R. Cabeza, R. Scherer, and G. Pfu rtscheller, “A fully on-line adaptive BCI,” IEEE Trans. Biomedical Engineering , vol. 53, no. 6, pp. 1214–1219, 2006

2006

[6] [6]

A comprehensive survey on transfer learning,

F. Zhuang, Z. Qi, K. Duan, D. Xi, Y . Zhu, H. Zhu, H. Xiong, an d Q. He, “A comprehensive survey on transfer learning,” Proc. of the IEEE , vol. 109, no. 1, pp. 43–76, 2021

2021

[7] [7]

A comprehensive survey on tes t-time adaptation under distribution shifts,

J. Liang, R. He, and T. Tan, “A comprehensive survey on tes t-time adaptation under distribution shifts,” Int’l Journal of Computer Vision , vol. 133, pp. 31–64, 2025

2025

[8] [8]

I n search of lost online test-time adaptation: A survey,

Z. Wang, Y . Luo, L. Zheng, Z. Chen, S. Wang, and Z. Huang, “I n search of lost online test-time adaptation: A survey,” Int’l Journal of Computer Vision, vol. 133, p. 1106–1139, 2024

2024