Pith. sign in

REVIEW 4 major objections 5 minor 91 references

Cortical-SSM outperforms attention- and convolution-based models at decoding imagined movements from EEG and ECoG signals.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 09:23 UTC pith:6BTCF6CP

load-bearing objection A genuinely new SSM-based MI decoder with informative ablations and honest interpretability work; the SOTA claim is plausible but not fully established because the closest SSM baselines are missing and baseline tuning is undocumented. the 4 major comments →

arxiv 2510.15371 v2 pith:6BTCF6CP submitted 2025-10-17 cs.CV cs.AI

Cortical-SSM: A Deep State Space Model for Motor Imagery Decoding from EEG Signals

Shuntaro Suzuki , Shunya Nagashima , Komei Sugiura This is my paper
classification cs.CV cs.AI
keywords motor imageryEEG decodingECoG decodingstate space modelsdeep SSMwavelet convolutionbrain-computer interfaceinterpretability
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to establish that a deep state space model, named Cortical-SSM, is a better and more interpretable alternative to Transformer- and CNN-based decoders for motor-imagery brain signals. It does so by explicitly modeling dependencies across three axes — temporal, spatial, and frequency — without compressing the time series into patches. On two large public EEG datasets and a clinical ECoG dataset, the authors report the best accuracy, macro-F1, AUROC, AUPRC, and Kappa among all compared methods, with gains over strong baselines such as EEG Conformer and EEGNet. The paper also claims that its visual explanations focus on neurophysiologically meaningful regions, such as the mu band and motor-cortex electrodes, which matters for clinical adoption of brain-computer interfaces.

Core claim

The central claim is that an architecture built on time-invariant, multi-input/multi-output state-space layers can decode motor imagery from EEG and ECoG more accurately than attention- and convolution-based baselines while remaining directly interpretable in the temporal, spatial, and frequency domains. Cortical-SSM processes each frequency component and each electrode separately through state-space layers, fed by a Wavelet-Convolution module that blends deterministic continuous-wavelet coefficients with learned convolutional features. In experiments, it reports the highest scores on every metric across all three benchmarks — for example, 81.62% accuracy versus 77.90% for EEG Conformer on O

What carries the argument

The load-bearing mechanism is a paired decomposition of the multichannel signal. A Wavelet-Convolution module combines a deterministic Morlet continuous-wavelet-transform filter bank with a trainable 1D convolution, both layer-normalized along time, to produce an M×F×T representation. Frequency-SSM then runs a time-invariant, multi-input/multi-output state-space layer independently over each frequency component to capture spatio-temporal dynamics; Channel-SSM runs the same type of layer independently over each electrode to capture temporal-frequency dynamics. The two outputs are average-pooled, concatenated, and classified. Because each branch keeps one axis explicit, the model avoids the te

Load-bearing premise

The central claim depends on the baseline comparisons being fair: the reported margins assume every baseline was tuned to a comparable degree and that eight fold-level paired observations are sufficient to support the statistical significance.

What would settle it

Re-run the three benchmarks with per-model hyperparameter search under a published training budget and report subject- or session-level paired statistics with more folds or nested cross-validation; if the accuracy gaps over EEG Conformer and EEGNet shrink to overlap or lose significance, the central outperformance claim collapses.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • If correct, state-space layers become a practical alternative to attention for motor-imagery decoding, offering linear-time long-sequence modeling without losing fine-grained temporal detail to patching.
  • The reported margins on ECoG-ALS, up to +9.61 accuracy over EEGNet, suggest particular value in invasive, lower-sample clinical settings.
  • The interpretability maps give clinicians a per-class, sample-agnostic view of which electrodes and frequency bands drive decisions, potentially easing validation and BCI illiteracy analysis.
  • The robustness to sequence length and SNR degradation reported in the appendices implies the architecture may hold up in real-world recordings with artifacts and variable trial durations.
  • Since the benchmarks use cross-subject and cross-session evaluation, the claimed gains are not limited to within-person calibration, which matters for practical deployment.

Where Pith is reading between the lines

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

  • The frequency/electrode factorization is generic: the same dual state-space design could be applied to other multivariate physiological arrays, such as EMG, ECG, or stereo EEG, with the interpretability maps serving as a hypothesis generator for which channels and bands carry task information.
  • The paper's own error analysis shows that Irrelevant Attention Error dominates, suggesting a direct testable extension: training the three branches jointly or progressively rather than independently should reduce overreliance on a single domain and improve accuracy.
  • A controlled comparison at identical sequence lengths between Cortical-SSM and a patched Transformer would isolate whether the gain comes from avoiding patchification or from the state-space backbone itself.
  • If the mu-band and C3/C4 attention maps are stable across subjects, they could be used for zero-shot transfer or subject-specific electrode selection in BCI calibration.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes Cortical-SSM, an S5-based deep state-space model for motor-imagery decoding from EEG and ECoG signals. The architecture couples a Wavelet-Convolution module (deterministic CWT features combined with a trainable 1D convolution) with two parallel SSM branches: Frequency-SSM models spatio-temporal dependencies per frequency component, and Channel-SSM models temporal-frequency dependencies per electrode. The method is evaluated on OpenBMI, Stieger2021, and a single-subject ECoG-ALS dataset using 8-fold cross-validation, with the claimed result that Cortical-SSM outperforms all included baselines on all reported metrics. The paper also provides ablations, visual explanations, sensitivity analyses, and an error analysis.

Significance. If the empirical claims are supported, the paper would make a useful contribution by showing that a MIMO, time-invariant SSM (S5) with explicit frequency/channel separation can be competitive with or better than convolutional and Transformer-based MI decoders, and that the proposed Wavelet-Convolution dual-branch feature extractor provides a favorable interpretability/accuracy trade-off. The manuscript contains several strengths: evaluation on three benchmarks including a clinical ECoG dataset, ablations of the main modules and of the temporal-kernel choice, sensitivity analyses with respect to sequence length and SNR, and an unusually candid error analysis. However, the central 'outperforms all baselines' claim is currently not fully supported because the baseline comparison is not verifiably fair, the statistical evidence is thin, and some dataset descriptions are internally inconsistent. The interpretability section also needs controls to avoid circularity. These issues are fixable, so I do not see grounds for rejection, but the paper needs substantial revision.

major comments (4)
  1. [§5.1/Table 1 vs. Appendix E.2] The SOTA claim rests on a fair baseline comparison, but Appendix E.2 reports a single training recipe (AdamW lr=1e-4, batch=8, 100 epochs) applied to every model. Baselines such as EEG Conformer, Medformer, PatchTST, iTransformer, and S-Mamba have architecture-specific optimal hyperparameters; without per-baseline tuning budgets or released code, the reported margins (+3.72 points on OpenBMI, +9.61 points on ECoG-ALS) may reflect under-tuned competitors. In addition, EEG-SSM, EEGMamba, and SWIM are cited in §2 but are absent from Tables 1, 5, and 6, so the claim of state-of-the-art among SSM-based EEG decoders is not actually tested. Please provide per-baseline hyperparameter settings, search budgets, and code, or narrow the claim accordingly.
  2. [Appendix E.1 (OpenBMI)] The dataset description is internally inconsistent. The text states that subjects performed 400 MI trials per session and that 'a total of 21,600 samples were collected from 54 subjects.' With two sessions per subject, the total should be 54 × 400 × 2 = 43,200 samples. The subsequent split uses 400 samples per subject (44×400=17,600 training, 5×400=2,000 validation, 5×400=2,000 test), which is consistent only if 400 is the total number of trials per subject, not per session. Please correct either the per-session description or the sample counts and split; this is load-bearing for the dataset description and reproducibility.
  3. [§5.1 (Wilcoxon tests)] The significance claim is based on paired Wilcoxon signed-rank tests on n=8 fold-level observations. With only 8 paired observations, the test has very low power, and no exact p-values or per-fold paired differences are reported. Moreover, no multiple-comparison correction is applied across roughly 18 baselines × 5 metrics. The Shapiro–Wilk normality check on n=8 is also too low-powered to justify the choice of nonparametric testing. To support 'statistically significant differences,' please report per-fold paired differences, exact p-values (including ties), effect sizes or confidence intervals, and state the multiple-comparison procedure; otherwise, the significance claims should be weakened.
  4. [§5.2 and Appendix D (interpretability)] The visual explanations are generated from features that are separated by construction into a frequency branch (U) and a channel branch (V), as shown in Eqs. (12)–(13). Observing that the explanations concentrate on the mu band and C3/C4 is therefore partly a consequence of the architecture, not an independent finding that the model exploits those features for the task. Additionally, Figures 3–4 average over successful cases only, which biases the interpretation. Please add control analyses—for example, comparison with a non-separated SSM baseline, permutation-based attribution, or attribution statistics over both correct and incorrect predictions—to support the interpretability claim.
minor comments (5)
  1. [Abstract vs. full text] The first abstract states that validation was performed on 'two large-scale public MI EEG datasets containing more than 50 subjects,' while the full-text abstract and §4 describe three benchmarks including the ECoG-ALS dataset. Please align the abstract with the full set of experiments.
  2. [Appendix E.1 (Stieger2021)] The Stieger2021 description is inconsistent: it first says 62 healthy subjects, then says 64 subjects, and later says 41 participants who completed all 11 sessions. Please clarify the exact number of subjects used and how the 41-participant subset is formed.
  3. [Tables 1, 5, 6] There are several typos: 'Corical-SSM' for Cortical-SSM, 'OpnBMI' for OpenBMI, 'PathcTST' for PatchTST, and 'iTransfromer' for iTransformer. Please proofread the tables and the corresponding text.
  4. [Eq. (3)] The frequency-bin indexing uses α=1...F, which does not include f_min. This is likely meant to be α=0...F−1 (or f_min + (α−1)(f_max−f_min)/F). Please correct or clarify.
  5. [Eq. (1)] The 1/2 and 1/2 fusion weights for CWT and Conv1D features are fixed without justification. The ablation in Table 2 shows that both branches help, but a sensitivity analysis over the fusion coefficient would strengthen the design choice.

Circularity Check

0 steps flagged

No circularity: the paper's claims are empirical evaluations on held-out benchmark folds, not derivations from fitted inputs.

full rationale

Cortical-SSM's central claim is that a proposed architecture obtains higher classification scores than baselines on three external MI benchmarks. No theoretical quantity is derived from the fitted parameters, no parameter is fitted to the evaluation data, and the reported numbers are direct cross-validated accuracies, F1, AUROC, AUPRC, and Kappa. The architecture uses standard external building blocks (CWT, S5, Grad-CAM), and the choice of S5 over Mamba is justified by external prior work, not by the paper's own conclusions. The interpretability analysis is post-hoc: Grad-CAM weights are learned, and the explanation maps could in principle highlight other frequency bands or electrodes, so agreement with the mu band and C3/C4 is an empirical observation rather than an equation-level reduction. The paper's limitations and Appendix F.3 explicitly acknowledge that independent domain processing may cause overreliance on a single domain, which weakens the interpretability claim but does not create circularity. The only self-citation (Kaneda et al., 2022) appears in Appendix A as a non-load-bearing example of time-series forecasting applications. The baseline-tuning and small-n Wilcoxon concerns are external validity risks, not circularity. Overall, the derivation chain is not circular; score 0.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claim rests on standard empirical assumptions plus several hand-chosen hyperparameters; no new physical or mathematical entities are introduced. The model is not derived from first principles, so the ledger mostly records design choices that are not independently motivated.

free parameters (5)
  • Wavelet frequency dimension F = 50
    Hand-chosen in Wavelet-Convolution (Appendix E.2); controls how many frequency components the CWT branch extracts.
  • CWT frequency range (fmin, fmax) = (1 Hz, 100 Hz)
    Chosen by hand; assumes all MI-relevant information lies in this band and that 50 scales resolve it.
  • Number of SSM blocks L = 2
    Hand-chosen for both Frequency-SSM and Channel-SSM; affects model capacity.
  • Conv1D kernel length K = f_sample/2 = 125
    Follows EEGNet, a design choice not derived from the task.
  • Training hyperparameters = lr=1e-4, batch=8, epochs=100
    Chosen without reported sensitivity analysis or per-baseline tuning budgets.
axioms (5)
  • domain assumption MI-relevant information is contained in the 1–100 Hz band and resolvable with F=50 CWT scales
    Appendix E.2 sets the Wavelet-Convolution frequency grid; if this band or resolution is wrong, the E-Branch loses discriminative content.
  • domain assumption Time-invariant MIMO S5 is more suitable for continuous EEG/ECoG than time-varying selection mechanisms
    Invoked in Section 3.3 with a citation to Gu & Dao 2024; not established by a task-specific comparison in this paper.
  • domain assumption Layer normalization along the temporal dimension avoids mutual noise in multivariate EEG/ECoG
    Justified in Section 3.2 by analogy to time-series forecasting literature; no ablation validates this choice for EEG.
  • domain assumption Known neurophysiological landmarks (mu band, C3/C4, hand-knob area) are the correct interpretability ground truth
    Used in Section 5.2 and Appendix F.3 to judge explanations; no quantitative correspondence measure is reported.
  • ad hoc to paper Fixed 1/2 and 1/2 fusion of CWT and Conv1D features is appropriate
    Equation 1 averages the two branches with equal weights; no learned gating or ablation over the fusion weight is provided.

pith-pipeline@v1.3.0-alltime-deepseek · 25095 in / 13557 out tokens · 111152 ms · 2026-08-04T09:23:43.267769+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Cortical-SSM: A Deep State Space Model for Motor Imagery Decoding from EEG Signals." pith.science (2026). https://pith.science/paper/6BTCF6CP

@misc{pith2026251015371,
  author       = {Pith},
  title        = {Pith review of: Cortical-SSM: A Deep State Space Model for Motor Imagery Decoding from EEG Signals},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6BTCF6CP}},
  note         = {Machine review of arXiv:2510.15371}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Classification of electroencephalogram (EEG) signals obtained during motor imagery (MI) has substantial application potential, including communication assistance and rehabilitation support for patients with motor impairments. These signals remain inherently susceptible to physiological artifacts (e.g., eye blinking and swallowing), which pose persistent challenges. Although Transformer-based approaches for classifying EEG signals have been widely adopted, they often struggle to capture fine-grained dependencies within them. To overcome these limitations, we propose Cortical-SSM, a novel architecture that extends deep state space models to capture integrated dependencies of EEG signals across temporal, spatial, and frequency domains. We validated our method across two large-scale public MI EEG datasets containing more than 50 subjects. Our method outperformed baseline methods on both benchmarks. Furthermore, visual explanations derived from our model indicate that it effectively captures neurophysiologically relevant regions of EEG signals. These results indicate that Cortical-SSM provides a robust and interpretable alternative to attention-based architectures for MI EEG decoding. By enabling physiologically grounded feature learning, our method advances the reliability of subject-independent EEG classification and supports the development of practical and clinically deployable brain-computer interface systems.

Figures

Figures reproduced from arXiv: 2510.15371 by Komei Sugiura, Shuntaro Suzuki, Shunya Nagashima.

Figure 1
Figure 1. Figure 1: Task overview. The input is EEG or ECoG signals recorded while the subject imagines [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed Cortical-SSM. Given an EEG or ECoG signal, the Wavelet [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Sample-agnostic visual ex￾planations of our proposed method on OpenBMI (Lee et al., 2019). Rows (a) and (b) display the temporal-frequency and spatio-temporal visual explanations for Subject 02, respectively. Columns (i) and (ii) correspond to left and right hand grasp, respectively. We further examine our proposed method by gener￾ating visual explanations, following the procedure de￾scribed in Appendix D … view at source ↗
Figure 4
Figure 4. Figure 4: Sample-agnostic visual explanations of our pro [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Electrode distribution in ECoG-ALS. Electrodes [PITH_FULL_IMAGE:figures/full_fig_p018_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Successful cases of the proposed method on OpenBMI (Lee et al., 2019). Rows (a) and (b) [PITH_FULL_IMAGE:figures/full_fig_p020_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Successful cases of the proposed method on ECoG-ALS. In this figure, Rows (a) and (b) [PITH_FULL_IMAGE:figures/full_fig_p021_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Failure cases of the proposed method. Rows (a) and (b) show samples from OpenBMI (Lee [PITH_FULL_IMAGE:figures/full_fig_p021_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Confusion matrices for OpenBMI (Lee et al., 2019) and ECoG-ALS. [PITH_FULL_IMAGE:figures/full_fig_p022_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Sensitivity analysis of EEG sequence length on OpenBMI (Lee et al., 2019). [PITH_FULL_IMAGE:figures/full_fig_p023_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Sensitivity analysis of EEG signal-to-noise ratio on OpenBMI (Lee et al., 2019). [PITH_FULL_IMAGE:figures/full_fig_p024_11.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

91 extracted references · 11 linked inside Pith

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Deep learning in EEG-based BCIs: a comprehensive review of transformer models, advantages, challenges, and applications

    Berdakh Abibullaev, Aigerim Keutayeva, and Amin Zollanvari. Deep learning in EEG-based BCIs: a comprehensive review of transformer models, advantages, challenges, and applications . IEEE Access, 11: 0 127271--127301, 2023

    2023

  3. [3]

    TimeMachine: A Time Series is Worth 4 Mambas for Long-term Forecasting

    Md Ahamed and Qiang Cheng. TimeMachine: A Time Series is Worth 4 Mambas for Long-term Forecasting . In ECAI, 2024

    2024

  4. [4]

    Deep learning techniques for classification of electroencephalogram (EEG) motor imagery (MI) signals: A review

    Hamdi Altaheri, Ghulam Muhammad, Mansour Alsulaiman, Syed Amin, Ghadir Ali Altuwaijri, et al. Deep learning techniques for classification of electroencephalogram (EEG) motor imagery (MI) signals: A review . Neural Computing and Applications, 35 0 (20): 0 14681--14722, 2023

    2023

  5. [5]

    Layer normalization

    Jimmy Lei Ba. Layer normalization . arXiv preprint arXiv:1607.06450, 2016

    Pith/arXiv arXiv 2016

  6. [6]

    wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations

    Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli. wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations . In NeurIPS, volume 33, pp.\ 12449--12460, 2020

    2020

  7. [7]

    Accurate medium-range global weather forecasting with 3D neural networks

    Kaifeng Bi, Lingxi Xie, Hengheng Zhang, Xin Chen, Xiaotao Gu, and Qi Tian. Accurate medium-range global weather forecasting with 3D neural networks . Nature, 619 0 (7970): 0 533--538, 2023

    2023

  8. [8]

    u ller, Dean Krusienski, Gerwin Schalk, Jonathan Wolpaw, Alois Schl \

    Benjamin Blankertz, Klaus-Robert M \"u ller, Dean Krusienski, Gerwin Schalk, Jonathan Wolpaw, Alois Schl \"o gl, Gert Pfurtscheller, Jos \'e del Mill \'a n, Michael Schr \"o der, and Niels Birbaumer. The BCI competition III: validating alternative approaches to actual BCI problems . TNSRE, 14 0 (2): 0 153--159, 2006

    2006

  9. [9]

    u ller-Putz, Alois Schl \

    Clemens Brunner, Robert Leeb, Gernot M \"u ller-Putz, Alois Schl \"o gl, and Gert Pfurtscheller. BCI Competition 2008--Graz data set A . Institute for knowledge discovery (laboratory of brain-computer interfaces), Graz University of Technology, 16: 0 1--6, 2008

    2008

  10. [10]

    EEG datasets for motor imagery brain--computer interface

    Hohyun Cho, Minkyu Ahn, Sangtae Ahn, Moonyoung Kwon, and Sung Chan Jun. EEG datasets for motor imagery brain--computer interface . GigaScience, 6 0 (7): 0 gix034, 2017

    2017

  11. [11]

    Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality

    Tri Dao and Albert Gu. Transformers are SSMs: Generalized Models and Efficient Algorithms Through Structured State Space Duality . In ICML, volume 235, pp.\ 10041--10071, 2024

    2024

  12. [12]

    EEG is better left alone

    Arnaud Delorme. EEG is better left alone . Scientific reports, 13 0 (1): 0 2372, 2023

    2023

  13. [13]

    Chris Dijkerman and Edward H

    H. Chris Dijkerman and Edward H. F. de Haan. Somatosensory processes subserving perception and action . Behavioral and Brain Sciences, 30 0 (2): 0 189–201, 2007

    2007

  14. [14]

    TSception:A Deep Learning Framework for Emotion Detection Using EEG

    Yi Ding, Neethu Robinson, Qiuhao Zeng, Duo Chen, Aung Aung Phyo Wai, Tih-Shih Lee, and Cuntai Guan. TSception:A Deep Learning Framework for Emotion Detection Using EEG . In IJCNN, pp.\ 1--7, 2020

    2020

  15. [15]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale . In ICLR, 2021

    2021

  16. [16]

    Wavelet Convolutional Neural Networks

    Shinetsu Fujieda, Kohei Takayama, and Toshiya Hachisuka. Wavelet Convolutional Neural Networks . ArXiv, abs/1805.08620, 2018

    Pith/arXiv arXiv 2018

  17. [17]

    Uni TS : A Unified Multi-Task Time Series Model

    Shanghua Gao, Teddy Koker, Owen Queen, Thomas Hartvigsen, Theodoros Tsiligkaridis, and Marinka Zitnik. Uni TS : A Unified Multi-Task Time Series Model . In NeurIPS, 2024

    2024

  18. [18]

    It’s Raw! Audio Generation with State-Space Models

    Karan Goel, Albert Gu, Chris Donahue, and Christopher R \'e . It’s Raw! Audio Generation with State-Space Models . In ICML, pp.\ 7616--7633, 2022

    2022

  19. [19]

    Mamba: Linear-Time Sequence Modeling with Selective State Spaces

    Albert Gu and Tri Dao. Mamba: Linear-Time Sequence Modeling with Selective State Spaces . In CoLM, 2024

    2024

  20. [20]

    HiPPO: Recurrent Memory with Optimal Polynomial Projections

    Albert Gu, Tri Dao, Stefano Ermon, Atri Rudra, and Christopher R \'e . HiPPO: Recurrent Memory with Optimal Polynomial Projections . In NeurIPS, volume 33, pp.\ 1474--1487, 2020

    2020

  21. [21]

    Combining Recurrent, Convolutional, and Continuous-time Models with Linear State-Space Layers

    Albert Gu, Isys Johnson, Karan Goel, Khaled Saab, Tri Dao, Atri Rudra, and Christopher R \'e . Combining Recurrent, Convolutional, and Continuous-time Models with Linear State-Space Layers . In NeurIPS, volume 34, pp.\ 572--585, 2021

    2021

  22. [22]

    Efficiently Modeling Long Sequences with Structured State Spaces

    Albert Gu, Karan Goel, and Christopher R\'e. Efficiently Modeling Long Sequences with Structured State Spaces . In ICLR, 2022

    2022

  23. [23]

    EEGMamba: Bidirectional State Space Model with Mixture of Experts for EEG Multi-task Classification

    Yiyu Gui, MingZhi Chen, Yuqi Su, Guibo Luo, and Yuchao Yang. EEGMamba: Bidirectional State Space Model with Mixture of Experts for EEG Multi-task Classification . In ICLR, 2025

    2025

  24. [24]

    MambaMorph: a Mamba-based Framework for Medical MR-CT Deformable Registration

    Tao Guo, Yinuo Wang, Shihao Shu, Diansheng Chen, Zhouping Tang, Cai Meng, and Xiangzhi Bai. MambaMorph: a Mamba-based Framework for Medical MR-CT Deformable Registration . arXiv preprint arXiv:2401.13934, 2024

    Pith/arXiv arXiv 2024

  25. [25]

    Flatten Transformer: Vision Transformer using Focused Linear Attention

    Dongchen Han, Xuran Pan, Yizeng Han, Shiji Song, and Gao Huang. Flatten Transformer: Vision Transformer using Focused Linear Attention . In ICCV, pp.\ 5961--5971, 2023

    2023

  26. [26]

    Physical principles of brain–computer interfaces and their applications for rehabilitation, robotics and control of human brain states

    Alexander Hramov, Vladimir Maksimenko, and Alexander Pisarchik. Physical principles of brain–computer interfaces and their applications for rehabilitation, robotics and control of human brain states . Physics Reports, 918: 0 1--133, 2021

    2021

  27. [27]

    Generative Learning for Financial Time Series with Irregular and Scale-Invariant Patterns

    Hongbin Huang, Minghua Chen, and Xiao Qiao. Generative Learning for Financial Time Series with Irregular and Scale-Invariant Patterns . In ICLR, 2024

    2024

  28. [28]

    Batch normalization: accelerating deep network training by reducing internal covariate shift

    Sergey Ioffe and Christian Szegedy. Batch normalization: accelerating deep network training by reducing internal covariate shift . In ICML, pp.\ 448–456, 2015

    2015

  29. [29]

    Large Brain Model for Learning Generic Representations with Tremendous EEG Data in BCI

    Weibang Jiang, Liming Zhao, and Bao liang Lu. Large Brain Model for Learning Generic Representations with Tremendous EEG Data in BCI . In ICLR, 2024

    2024

  30. [30]

    Dual-path Mamba: Short and Long-term Bidirectional Selective Structured State Space Models for Speech Separation

    Xilin Jiang, Cong Han, and Nima Mesgarani. Dual-path Mamba: Short and Long-term Bidirectional Selective Structured State Space Models for Speech Separation . In ICASSP, pp.\ 1--5, 2025

    2025

  31. [31]

    R. Kalman. A New Approach to Linear Filtering and Prediction Problems . Journal of Basic Engineering, 82 0 (1): 0 35--45, 1960

    1960

  32. [32]

    Flare Transformer: Solar Flare Prediction using Magnetograms and Sunspot Physical Features

    Kanta Kaneda, Yuiga Wada, Tsumugi Iida, Naoto Nishizuka, Y\^uki Kubo, and Komei Sugiura. Flare Transformer: Solar Flare Prediction using Magnetograms and Sunspot Physical Features . In ACCV, pp.\ 1488--1503, 2022

    2022

  33. [33]

    A large electroencephalographic motor imagery dataset for electroencephalographic brain computer interfaces

    Murat Kaya, Mustafa Binli, Erkan Ozbay, Hilmi Yanar, and Yuriy Mishchenko. A large electroencephalographic motor imagery dataset for electroencephalographic brain computer interfaces . Scientific data, 5 0 (1): 0 1--16, 2018

    2018

  34. [34]

    Reversible Instance Normalization for Accurate Time-Series Forecasting against Distribution Shift

    Taesung Kim, Jinhee Kim, Yunwon Tae, Cheonbok Park, Jang-Ho Choi, and Jaegul Choo. Reversible Instance Normalization for Accurate Time-Series Forecasting against Distribution Shift . In ICLR, 2022

    2022

  35. [35]

    The ten-twenty electrode system of the international federation

    George Klem. The ten-twenty electrode system of the international federation. The international federation of clinical neurophysiology . Electroencephalogr. Clin. Neurophysiol. Suppl., 52: 0 3--6, 1999

    1999

  36. [36]

    An experimental review on deep learning architectures for time series forecasting

    Pedro Lara-Ben \' tez, Manuel Carranza-Garc \' a, and Jos \'e C Riquelme. An experimental review on deep learning architectures for time series forecasting . International journal of neural systems, 31 0 (03): 0 2130001, 2021

    2021

  37. [37]

    EEGNet: A Compact Convolutional Neural Network for EEG-based Brain--Computer Interfaces

    Vernon Lawhern, Amelia Solon, Nicholas Waytowich, Stephen Gordon, Chou Hung, and Brent Lance. EEGNet: A Compact Convolutional Neural Network for EEG-based Brain--Computer Interfaces . Journal of Neural Engineering , 15 0 (5): 0 056013, 2018

    2018

  38. [38]

    EEG dataset and OpenBMI toolbox for three BCI paradigms: an investigation into BCI illiteracy

    Min-Ho Lee, O-Yeon Kwon, Yong-Jeong Kim, Hong-Kyung Kim, Young-Eun Lee, John Williamson, Siamac Fazli, and Seong-Whan Lee. EEG dataset and OpenBMI toolbox for three BCI paradigms: an investigation into BCI illiteracy . GigaScience, 8 0 (5): 0 giz002, 2019

    2019

  39. [39]

    Frozen Language Model Helps ECG Zero-Shot Learning

    Jun Li, Che Liu, Sibo Cheng, Rossella Arcucci, and Shenda Hong. Frozen Language Model Helps ECG Zero-Shot Learning . In MIDL, pp.\ 402--415, 2024 a

    2024

  40. [40]

    SPMamba: State-space model is all you need in speech separation

    Kai Li, Guo Chen, Runxuan Yang, and Xiaolin Hu. SPMamba: State-space model is all you need in speech separation . arXiv preprint arXiv:2404.02063, 2024 b

    Pith/arXiv arXiv 2024

  41. [41]

    Foundation models for time series analysis: A tutorial and survey

    Yuxuan Liang, Haomin Wen, Yuqi Nie, Yushan Jiang, Ming Jin, Dongjin Song, Shirui Pan, and Qingsong Wen. Foundation models for time series analysis: A tutorial and survey . In KDD, pp.\ 6555--6565, 2024

    2024

  42. [42]

    Multi-Level Wavelet Convolutional Neural Networks

    Pengju Liu, Hongzhi Zhang, Wei Lian, and Wangmeng Zuo. Multi-Level Wavelet Convolutional Neural Networks . IEEE Access, 7: 0 74973--74985, 2019

    2019

  43. [43]

    Non-stationary transformers: Exploring the stationarity in time series forecasting

    Yong Liu, Haixu Wu, Jianmin Wang, and Mingsheng Long. Non-stationary transformers: Exploring the stationarity in time series forecasting . In NeurIPS, volume 35, pp.\ 9881--9893, 2022

    2022

  44. [44]

    iTransformer: Inverted Transformers Are Effective for Time Series Forecasting

    Yong Liu, Tengge Hu, Haoran Zhang, Haixu Wu, Shiyu Wang, Lintao Ma, and Mingsheng Long. iTransformer: Inverted Transformers Are Effective for Time Series Forecasting . In ICLR, 2024

    2024

  45. [45]

    U-Mamba: Enhancing Long-Range Dependency for Biomedical Image Segmentation

    Jun Ma, Feifei Li, and Bo Wang. U-Mamba: Enhancing Long-Range Dependency for Biomedical Image Segmentation . arXiv preprint arXiv:2401.04722, 2024

    Pith/arXiv arXiv 2024

  46. [46]

    Mega: Moving Average Equipped Gated Attention

    Xuezhe Ma, Chunting Zhou, Xiang Kong, Junxian He, Liangke Gui, Graham Neubig, Jonathan May, and Zettlemoyer Luke. Mega: Moving Average Equipped Gated Attention . arXiv preprint arXiv:2209.10655, 2022

    Pith/arXiv arXiv 2022

  47. [47]

    FBCNet: A Multi-view Convolutional Neural Network for Brain-Computer Interface

    Ravikiran Mane, Effie Chew, Karen Chua, Kai Keng Ang, Neethu Robinson, A Prasad Vinod, Seong-Whan Lee, et al. FBCNet: A Multi-view Convolutional Neural Network for Brain-Computer Interface . arXiv preprint arXiv:2104.01233, 2021

    Pith/arXiv arXiv 2021

  48. [48]

    Statistical Evaluation of Factors Influencing Inter-Session and Inter-Subject Variability in EEG-Based Brain Computer Interface

    Rito Clifford Maswanganyi, Chunling Tu, Pius Adewale Owolawi, and Shengzhi Du. Statistical Evaluation of Factors Influencing Inter-Session and Inter-Subject Variability in EEG-Based Brain Computer Interface . IEEE Access, 10: 0 96821--96839, 2022

    2022

  49. [49]

    LMDA-Net: A lightweight multi-dimensional attention network for general EEG-based brain-computer interfaces and interpretability

    Zhengqing Miao, Meirong Zhao, Xin Zhang, and Dong Ming. LMDA-Net: A lightweight multi-dimensional attention network for general EEG-based brain-computer interfaces and interpretability . NeuroImage, 276: 0 120209, 2023

    2023

  50. [50]

    Spectral changes in cortical surface potentials during motor movement

    J Miller, C Leuthardt, Gerwin Schalk, PN Rao, R Anderson, W Moran, W Miller, and G Ojemann. Spectral changes in cortical surface potentials during motor movement . Journal of Neuroscience, 27 0 (9): 0 2424--2432, 2007

    2007

  51. [51]

    S4ND: Modeling Images and Videos as Multidimensional Signals Using State Spaces

    Eric Nguyen, Karan Goel, Albert Gu, Gordon Downs, Preey Shah, Tri Dao, Stephen Baccus, and Christopher R\'e. S4ND: Modeling Images and Videos as Multidimensional Signals Using State Spaces . In NeurIPS, volume 35, pp.\ 2846--2861, 2022

    2022

  52. [52]

    A Time Series is Worth 64 Words: Long-term Forecasting with Transformers

    Yuqi Nie, Nam Nguyen, Phanwadee Sinthong, and Jayant Kalagnanam. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers . In ICLR, 2023

    2023

  53. [53]

    10-10 electrode system for EEG recording

    Marc Nuwer. 10-10 electrode system for EEG recording . Clin. Neurophysiol., 129 0 (5): 0 1103--1103, 2018

    2018

  54. [55]

    Sensorimotor disturbances in patients with lesions of the parietal cortex

    Max Pause, Erwin Kunesch, Ferdinand Binkofski, and Hans-Joachim Freund. Sensorimotor disturbances in patients with lesions of the parietal cortex . Brain, 112 0 (6): 0 1599--1625, 1989

    1989

  55. [56]

    Motor imagery and direct brain-computer communication

    Gert Pfurtscheller and Christa Neuper. Motor imagery and direct brain-computer communication . Proceedings of the IEEE, 89 0 (7): 0 1123--1134, 2001

    2001

  56. [57]

    Mu rhythm (de) synchronization and EEG single-trial classification of different motor imagery tasks

    Gert Pfurtscheller, Clemens Brunner, Alois Schl \"o gl, et al. Mu rhythm (de) synchronization and EEG single-trial classification of different motor imagery tasks . NeuroImage, 31 0 (1): 0 153--159, 2006

    2006

  57. [58]

    Brain computer interface: control signals review

    Rabie Ramadan and Athanasios Vasilakos. Brain computer interface: control signals review . Neurocomputing, 223: 0 26--44, 2017

    2017

  58. [59]

    Combiner: Full Attention Transformer with Sparse Computation Cost

    Hongyu Ren, Hanjun Dai, Zihang Dai, Mengjiao Yang, Jure Leskovec, Dale Schuurmans, and Bo Dai. Combiner: Full Attention Transformer with Sparse Computation Cost . In NeurIPS, volume 34, pp.\ 22470--22482, 2021

    2021

  59. [60]

    BCI2000: a general-purpose brain-computer interface (BCI) system

    Gerwin Schalk, Dennis McFarland, Thilo Hinterberger, Niels Birbaumer, and Jonathan Wolpaw. BCI2000: a general-purpose brain-computer interface (BCI) system . TBME, 51 0 (6): 0 1034--1043, 2004

    2004

  60. [61]

    Deep learning with convolutional neural networks for EEG decoding and visualization

    Robin Schirrmeister, Jost Springenberg, Lukas Fiederer, Martin Glasstetter, Katharina Eggensperger, Michael Tangermann, et al. Deep learning with convolutional neural networks for EEG decoding and visualization . Human brain mapping, 38 0 (11): 0 5391--5420, 2017

    2017

  61. [62]

    Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra

    Ramprasaath R. Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization . In ICCV, pp.\ 618--626, 2017

    2017

  62. [63]

    Simplified State Space Layers for Sequence Modeling

    Jimmy Smith, Andrew Warrington, and Scott Linderman. Simplified State Space Layers for Sequence Modeling . In ICLR, 2023

    2023

  63. [64]

    EEG Conformer: Convolutional Transformer for EEG Decoding and Visualization

    Yonghao Song, Qingqing Zheng, Bingchuan Liu, and Xiaorong Gao. EEG Conformer: Convolutional Transformer for EEG Decoding and Visualization . TNSRE, 31: 0 710--719, 2022

    2022

  64. [65]

    S7: Selective and Simplified State Space Layers for Sequence Modeling

    Taylan Soydan, Nikola Zubi \'c , Nico Messikommer, Siddhartha Mishra, and Davide Scaramuzza. S7: Selective and Simplified State Space Layers for Sequence Modeling . arXiv preprint arXiv:2410.03464, 2024

    Pith/arXiv arXiv 2024

  65. [66]

    Continuous sensorimotor rhythm based brain computer interface learning in a large population

    James Stieger, Stephen Engel, and Bin He. Continuous sensorimotor rhythm based brain computer interface learning in a large population . Scientific Data, 8 0 (1): 0 98, 2021

    2021

  66. [67]

    A practical guide to wavelet analysis

    Christopher Torrence and Gilbert Compo. A practical guide to wavelet analysis . BAMS, 79 0 (1): 0 61--78, 1998

    1998

  67. [68]

    EEG-SSM: Leveraging State-Space Model for Dementia Detection

    Xuan-The Tran, Linh Le, Quoc Toan Nguyen, Thomas Do, and Chin-Teng Lin. EEG-SSM: Leveraging State-Space Model for Dementia Detection . arXiv preprint arXiv:2407.17801, 2024

    Pith/arXiv arXiv 2024

  68. [69]

    Attention is All you Need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, ukasz Kaiser, and Illia Polosukhin. Attention is All you Need . In NeurIPS, volume 30, 2017

    2017

  69. [70]

    Mamba-Reg: Vision Mamba Also Needs Registers

    Feng Wang, Jiahao Wang, Sucheng Ren, Guoyizhe Wei, Jieru Mei, Wei Shao, Yuyin Zhou, Alan Yuille, and Cihang Xie. Mamba-Reg: Vision Mamba Also Needs Registers . In ICLR, 2025 a

    2025

  70. [71]

    EEGPT : Pretrained transformer for universal and reliable representation of EEG signals

    Guangyu Wang, Wenchao Liu, Yuhong He, Cong Xu, Lin Ma, and Haifeng Li. EEGPT : Pretrained transformer for universal and reliable representation of EEG signals. In NeurIPS, volume 37, pp.\ 39249--39280, 2024 a

    2024

  71. [72]

    TimeMixer: Decomposable Multiscale Mixing for Time Series Forecasting

    Shiyu Wang, Haixu Wu, Xiaoming Shi, Tengge Hu, Huakun Luo, Lintao Ma, James Zhang, and Jun Zhou. TimeMixer: Decomposable Multiscale Mixing for Time Series Forecasting . In ICLR, 2024 b

    2024

  72. [73]

    State Space Model for New-Generation Network Alternative to Transformers: A Survey

    Xiao Wang, Shiao Wang, Yuhe Ding, Yuehang Li, Wentao Wu, Yao Rong, Weizhe Kong, Ju Huang, Shihao Li, Haoxiang Yang, et al. State Space Model for New-Generation Network Alternative to Transformers: A Survey . arXiv preprint arXiv:2404.09516, 2024 c

    Pith/arXiv arXiv 2024

  73. [74]

    Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification

    Yihe Wang, Nan Huang, Taida Li, Yujun Yan, and Xiang Zhang. Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification . In NeurIPS, volume 37, pp.\ 36314--36341, 2024 d

    2024

  74. [75]

    Is Mamba Effective for Time Series Forecasting? Neurocomputing, 619: 0 129178, 2025 b

    Zihan Wang, Fanheng Kong, Shi Feng, Ming Wang, Xiaocui Yang, Han Zhao, Daling Wang, and Yifei Zhang. Is Mamba Effective for Time Series Forecasting? Neurocomputing, 619: 0 129178, 2025 b

    2025

  75. [76]

    Wavelet Pooling for Convolutional Neural Networks

    Travis Williams and Robert Li. Wavelet Pooling for Convolutional Neural Networks . In ICLR, 2018

    2018

  76. [77]

    Brain-computer interfaces

    Jonathan Wolpaw. Brain-computer interfaces . Handbook of Clinical Neurology, 110: 0 67--74, 2013

    2013

  77. [78]

    Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting

    Haixu Wu, Jiehui Xu, Jianmin Wang, and Mingsheng Long. Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting . In NeurIPS, volume 34, pp.\ 22419--22430, 2021

    2021

  78. [79]

    TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis

    Haixu Wu, Tengge Hu, Yong Liu, Hang Zhou, Jianmin Wang, and Mingsheng Long. TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis . In ICLR, 2023

    2023

  79. [80]

    Group normalization

    Yuxin Wu and Kaiming He. Group normalization . In ECCV, pp.\ 3--19, 2018

    2018

  80. [81]

    Localization of the motor hand area to a knob on the precentral gyrus

    A Yousry, D Schmid, H Alkadhi, D Schmidt, A Peraud, A Buettner, and P Winkler. Localization of the motor hand area to a knob on the precentral gyrus. A new landmark. Brain, 120 0 (1): 0 141--157, 1997

    1997

Showing first 80 references.