pith. sign in

arxiv: 2606.25456 · v1 · pith:TVNQVUZ4new · submitted 2026-06-24 · 💻 cs.LG

Towards Robust EEG Decoding Based on Riemannian Self-Attention

Shaocheng Jin , Tao Zhou , Rui Wang , Ziheng Chen , Xiaoqing Luo , Xiaojun Wu , Josef Kittler This is my paper

Pith reviewed 2026-06-25 21:33 UTC · model grok-4.3

classification 💻 cs.LG
keywords EEG decodingRiemannian manifoldBures-Wasserstein metricself-attentionbrain-computer interfaceSPD matricesdeep learning
0
0 comments X

The pith

A Riemannian self-attention network on the Bures-Wasserstein metric improves robustness of EEG decoding for brain-computer interfaces.

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

The paper introduces a Riemannian self-attention network for EEG decoding that operates on symmetric positive definite matrices using the Bures-Wasserstein metric instead of the conventional affine-invariant metric. This choice addresses the quadratic computational cost and sensitivity to ill-conditioned matrices that arise with noisy EEG data. The model is extended to a learnable generalized version that incorporates power deformation of the metric. Experimental validation on three standard EEG datasets demonstrates improved robustness. Readers interested in brain-computer interfaces would care because better handling of low signal-to-noise ratios could expand practical BCI use in rehabilitation and assistive devices.

Core claim

We propose a Riemannian self-attention network based on the Bures-Wasserstein metric for SPD learning in EEG decoding. We further extend it to a learnable power-deformed generalized Bures-Wasserstein version called GBWAtt to provide a more nuanced representation of the SPD manifold. This overcomes limitations of basic architectures that fail to capture local signal relationships and of the affine-invariant metric that has quadratic dependency and issues with ill-conditioned matrices.

What carries the argument

The Bures-Wasserstein metric (BWM) and its power-deformed generalized form (GBW) within a self-attention network, which provides linear dependence on SPD matrices and enables learnable geometric structure for EEG covariance matrices.

If this is right

  • The network can handle ill-conditioned SPD matrices arising from low-SNR EEG without breakdown.
  • Local relationships between EEG signals are explicitly modeled via self-attention on the manifold.
  • Decoding performance improves across multiple benchmarking datasets for BCI tasks.
  • Learnable metric deformation allows adaptation to the specific geometric properties of the data.

Where Pith is reading between the lines

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

  • If the metric choice proves superior, it may encourage broader adoption of the Bures-Wasserstein metric in other manifold learning tasks involving noisy covariance data.
  • The learnable extension suggests that metric parameters could be optimized jointly with the network for subject-specific EEG patterns.
  • This architecture might inspire similar attention mechanisms on other Riemannian manifolds used in signal processing.

Load-bearing premise

The Bures-Wasserstein metric supplies a superior geometric representation of SPD matrices for low-SNR EEG signals compared with the affine-invariant metric, and the added self-attention layer meaningfully captures local signal relationships that basic architectures miss.

What would settle it

A direct comparison showing no improvement in accuracy or robustness when replacing the affine-invariant metric with the Bures-Wasserstein metric inside the self-attention architecture on the three EEG benchmarking datasets would falsify the central claim.

Figures

Figures reproduced from arXiv: 2606.25456 by Josef Kittler, Rui Wang, Shaocheng Jin, Tao Zhou, Xiaojun Wu, Xiaoqing Luo, Ziheng Chen.

Figure 1
Figure 1. Figure 1: An overview of the proposed BWAtt. (a) illustrates the overall architecture of the BWAtt network; (b) shows the SPD [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of the limited data on the MAMEM. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Heatmaps of GBWAtt for the S3 subject across four motor-imagery classes on the BCIC-IV-2a dataset. The x-axis and [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The spatial topo-maps and the diagram of electrode distribution of GBWAtt for the S3 subject across four motor [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Heatmaps of GBWAtt for the S11 subject across five different frequencies on the MAMEM-SSVEP-II dataset. The x-axis [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The spatial topo-maps and the diagram of electrode distribution of GBWAtt for the S11 subject across five different [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The heatmaps and the visualization result of GBWAtt for two classes on the BCI-ERN datasets S7 model. The x-axis [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
read the original abstract

Brain-Computer Interface (BCI) based on electroencephalography (EEG) enables direct interaction between the brain and external environments and has significant applications in assistive technologies, medical rehabilitation, and entertainment. Recently, EEG decoding methods based on Symmetric Positive Definite (SPD) learning have demonstrated superior performance. However, these methods typically employ basic network architectures and do not explicitly capture local relationships between EEG signals. This limitation is problematic for EEG signals due to their inherently low Signal-to-Noise Ratio (SNR). Moreover, most existing Riemannian manifold-based methods are restricted to specific metrics. The most widely used is the Affine-Invariant Metric (AIM). However, it has a quadratic dependency on the SPD matrices and cannot handle ill-conditioned SPD matrices, which hinders the effectiveness of networks. In contrast, the Bures-Wasserstein Metric (BWM) exhibits linear dependence on SPD matrices and demonstrates superior performance for ill conditioning. To overcome these challenges, we propose a Riemannian self-attention network based on the BWM. Additionally, the recently introduced power-deformed generalized Bures-Wasserstein metric reveals a nonlinear relationship between SPD matrices and matrix power deformation. This metric provides a more nuanced representation of the geometric structure of the SPD manifold. Consequently, we extend our model to a learnable version. For simplicity, we refer to it as GBWAtt. Experimental results on three EEG benchmarking datasets validate the robustness and effectiveness of our proposed method. The code is available at https://github.com/jissc/GBWAtt.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 3 minor

Summary. The paper proposes GBWAtt, a Riemannian self-attention network for EEG decoding that operates on SPD matrices using the Bures-Wasserstein metric (BWM) and its recently introduced power-deformed generalization. It argues that BWM provides linear dependence on the matrices and better handling of ill-conditioned cases than the affine-invariant metric (AIM), while the self-attention layer captures local signal relationships missed by prior Riemannian networks; the learnable variant optimizes the power deformation parameter. Experimental results on three EEG benchmarking datasets are presented to support robustness and effectiveness, with code released at the provided GitHub link.

Significance. If the performance claims hold, the work offers a practical advance for low-SNR EEG decoding in BCI by replacing AIM with a metric that avoids quadratic scaling and ill-conditioning issues while adding attention for local structure. Explicit release of code is a clear strength that enables direct reproducibility and extension.

major comments (1)
  1. [Method (learnable GBW extension) and Experiments] The learnable power deformation parameter is optimized on the same three benchmark datasets used to claim superiority over fixed-metric baselines; without an explicit statement of the optimization protocol (e.g., nested cross-validation or held-out validation splits) this introduces a circularity risk that directly affects the central empirical claim.
minor comments (3)
  1. [Abstract and §3] The abstract states that the model 'extends our model to a learnable version' but does not define the exact parameterization of the power deformation inside the attention module; an equation or pseudocode block would clarify the implementation.
  2. [Experiments] Table or figure captions for the three datasets should explicitly list the number of subjects, trials, and classes to allow immediate comparison with prior Riemannian EEG work.
  3. [Introduction] The claim that BWM 'demonstrates superior performance for ill conditioning' would benefit from a short reference to the specific prior result or a small illustrative matrix example in the text.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive evaluation and the constructive comment. We address the concern regarding the learnable parameter optimization protocol below and commit to a revision that provides the requested transparency.

read point-by-point responses

  1. Referee: [Method (learnable GBW extension) and Experiments] The learnable power deformation parameter is optimized on the same three benchmark datasets used to claim superiority over fixed-metric baselines; without an explicit statement of the optimization protocol (e.g., nested cross-validation or held-out validation splits) this introduces a circularity risk that directly affects the central empirical claim.

    Authors: We agree that the absence of an explicit optimization protocol description creates a legitimate concern about potential circularity. The revised manuscript will add a dedicated subsection under Experiments that details the protocol: the power parameter is tuned exclusively via inner-loop cross-validation on the training folds of each outer evaluation split, with no access to test data. This ensures the reported superiority of the learnable GBWAtt variant is evaluated on held-out data. We will also report the specific search ranges and selected values per dataset for full reproducibility. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper proposes a Riemannian self-attention network using the Bures-Wasserstein metric (drawn from prior literature) and its power-deformed generalization extended to a learnable form, with effectiveness shown via empirical results on three EEG benchmarks. No load-bearing step reduces by the paper's equations or self-citation to a self-definition, fitted parameter renamed as prediction, or imported uniqueness theorem. The central claims rest on architectural design plus external validation rather than an internal derivation that collapses to its inputs by construction.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The claim rests on the domain assumption that EEG covariances are well-modeled as SPD matrices, the geometric properties asserted for BWM versus AIM, and one learnable scalar in the generalized metric; no new physical entities are postulated.

free parameters (1)
  • power deformation parameter
    Learnable scalar introduced to deform the generalized Bures-Wasserstein metric during training.
axioms (2)
  • domain assumption EEG signals are appropriately summarized by SPD covariance matrices on a Riemannian manifold
    Invoked as the starting point for all Riemannian EEG methods.
  • domain assumption BWM exhibits linear dependence on SPD matrices and superior behavior on ill-conditioned cases relative to AIM
    Stated as established fact motivating the architecture choice.

pith-pipeline@v0.9.1-grok · 5817 in / 1371 out tokens · 27511 ms · 2026-06-25T21:33:57.271303+00:00 · methodology

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

76 extracted references · 2 linked inside Pith

  1. [1]

    Ghadir Ali Altuwaijri, Ghulam Muhammad, Hamdi Altaheri, and Mansour Alsu- laiman. 2022. A multi-branch convolutional neural network with squeeze-and- excitation attention blocks for EEG-based motor imagery signals classification. Diagnostics(2022), 995

    2022

  2. [2]

    Sion An, Soopil Kim, Philip Chikontwe, and Sang Hyun Park. 2023. Dual attention relation network with fine-tuning for few-shot EEG motor imagery classification. IEEE Trans. Neural Netw. Learn. Syst.(2023), 15479 – 15493

    2023

  3. [3]

    Vincent Arsigny, Pierre Fillard, Xavier Pennec, and Nicholas Ayache. 2007. Geo- metric means in a novel vector space structure on symmetric positive-definite matrices.SIAM J. Matrix Anal. Appl.(2007), 328–347

    2007

  4. [5]

    InInternational confer- ence on latent variable analysis and signal separation

    Riemannian geometry applied to BCI classification. InInternational confer- ence on latent variable analysis and signal separation

  5. [6]

    Alexandre Barachant, Stéphane Bonnet, Marco Congedo, and Christian Jutten

  6. [7]

    Classification of covariance matrices using a Riemannian-based kernel for BCI applications.Neurocomputing(2013), 172–178

    2013

  7. [8]

    Rajendra Bhatia, Tanvi Jain, and Yongdo Lim. 2019. On the Bures–Wasserstein distance between positive definite matrices.Expo. Math.(2019), 165–191

    2019

  8. [9]

    Joyce Chelangat Bore, Peiyang Li, Lin Jiang, Walid MA Ayedh, Chunli Chen, Dennis Joe Harmah, Dezhong Yao, Zehong Cao, and Peng Xu. 2021. A long short-term memory network for sparse spatiotemporal EEG source imaging. IEEE Trans. Med. Imaging.(2021), 3787–3800

    2021

  9. [10]

    Daniel Brooks, Olivier Schwander, Frédéric Barbaresco, Jean-Yves Schneider, and Matthieu Cord. 2019. Riemannian batch normalization for SPD neural networks. InNeurIPS

    2019

  10. [11]

    Clemens Brunner, Robert Leeb, Gernot Müller-Putz, Alois Schlögl, and Gert Pfurtscheller. 2008. BCI Competition 2008–Graz data set A.Graz University of Technology, Austria(2008)

    2008

  11. [12]

    Rudrasis Chakraborty, Jose Bouza, Jonathan Manton, and Baba C Vemuri. 2022. Manifoldnet: A deep neural network for manifold-valued data with applications. IEEE Trans. Pattern Anal. Mach. Intell.(2022), 799–810

    2022

  12. [13]

    Rudrasis Chakraborty, Chun-Hao Yang, Xingjian Zhen, Monami Banerjee, Derek Archer, David Vaillancourt, Vikas Singh, and Baba Vemuri. 2018. A statistical recurrent model on the manifold of symmetric positive definite matrices. In NeurIPS

    2018

  13. [14]

    Ziheng Chen, Yue Song, Gaowen Liu, Ramana Rao Kompella, Xiao-Jun Wu, and Nicu Sebe. 2023. Riemannian Multiclass Logistics Regression for SPD Neural Networks.arXiv preprint arXiv:2305.11288(2023)

    arXiv 2023

  14. [15]

    Ziheng Chen, Yue Song, Yunmei Liu, and Nicu Sebe. 2024. A Lie Group Approach to Riemannian Normalization for SPD Neural Networks. InICLR

    2024

  15. [16]

    Ziheng Chen, Yue Song, Xiaojun Wu, Gaowen Liu, and Nicu Sebe. 2025. Under- standing Matrix Function Normalizations in Covariance Pooling through the Lens of Riemannian Geometry. InICLR

    2025

  16. [17]

    Ziheng Chen, Tianyang Xu, Xiao-Jun Wu, Rui Wang, Zhiwu Huang, and Josef Kittler. 2023. Riemannian local mechanism for SPD neural networks. InAAAI

    2023

  17. [18]

    Ziheng Chen, Tianyang Xu, Xiao-Jun Wu, Rui Wang, Zhiwu Huang, and Josef Kittler. 2023. Riemannian Local Mechanism for SPD Neural Networks. InAAAI

    2023

  18. [19]

    Marco Congedo, Alexandre Barachant, and Rajendra Bhatia. 2017. Riemannian geometry for EEG-based brain-computer interfaces; a primer and a review.Brain Comput. Interfaces(2017), 155–174

    2017

  19. [20]

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xi- aohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. 2020. An image is worth 16x16 words: Transformers for image recognition at scale.arXiv preprint arXiv:2010.11929(2020)

    Pith/arXiv arXiv 2020

  20. [21]

    Zhi Gao, Yuwei Wu, Mehrtash Harandi, and Yunde Jia. 2020. A Robust Distance Measure for Similarity-Based Classification on the SPD Manifold.IEEE Trans. Neural Netw. Learn. Syst.(2020), 3230–3244

    2020

  21. [22]

    Greg Hajcak. 2012. What we’ve learned from mistakes: Insights from error-related brain activity.Curr. Dir. Psychol.(2012)

    2012

  22. [23]

    Andi Han, Bamdev Mishra, Pratik Jawanpuria, and Junbin Gao. 2021. General- ized Bures-Wasserstein geometry for positive definite matrices.arXiv preprint arXiv:2110.10464(2021)

    arXiv 2021

  23. [24]

    Andi Han, Bamdev Mishra, Pratik Jawanpuria, and Junbin Gao. 2023. Learning with symmetric positive definite matrices via generalized Bures-Wasserstein geometry. InInternational Conference on Geometric Science of Information

    2023

  24. [25]

    Andi Han, Bamdev Mishra, Pratik Kumar Jawanpuria, and Junbin Gao. 2021. On Riemannian optimization over positive definite matrices with the Bures- Wasserstein geometry. InNeurIPS

    2021

  25. [26]

    Chengcheng Han, Guanghua Xu, Jun Xie, Chaoyang Chen, and Sicong Zhang

  26. [27]

    Highly interactive brain–computer interface based on flicker-free steady- state motion visual evoked potential.Scientific reports(2018), 5835

    2018

  27. [28]

    Mehrtash Harandi and Mathieu Salzmann. 2015. Riemannian coding and dictio- nary learning: Kernels to the rescue. InCVPR

    2015

  28. [29]

    Mehrtash Harandi, Mathieu Salzmann, and Richard Hartley. 2018. Dimensionality reduction on SPD manifolds: The emergence of geometry-aware methods.IEEE Trans. Pattern Anal. Mach. Intell.(2018), 48–62

    2018

  29. [30]

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recognition. InCVPR

    2016

  30. [31]

    Christoph S Herrmann. 2001. Human EEG responses to 1–100 Hz flicker: reso- nance phenomena in visual cortex and their potential correlation to cognitive phenomena.Exp. Brain Res.(2001), 346–353

    2001

  31. [32]

    Andrew G Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, and Hartwig Adam. 2017. Mobilenets: Efficient convolutional neural networks for mobile vision applications.arXiv preprint arXiv:1704.04861(2017)

    Pith/arXiv arXiv 2017

  32. [33]

    Chen Hu, Ziheng Chen, Rui Wang, Yefeng Zheng, and Nicu Sebe. 2026. Riemann- ian High-Order Pooling for Brain Foundation Models. InICLR

    2026

  33. [34]

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger

  34. [35]

    Densely connected convolutional networks. InCVPR

  35. [36]

    Zhiwu Huang and Luc Van Gool. 2017. A Riemannian network for spd matrix learning. InAAAI

    2017

  36. [37]

    Zhiwu Huang, Ruiping Wang, Shiguang Shan, Xianqiu Li, and Xilin Chen. 2015. Log-euclidean metric learning on symmetric positive definite manifold with application to image set classification. InICML

    2015

  37. [38]

    Wei-Bang Jiang, Liming Zhao, and Bao-Liang Lu. 2024. Large brain model for learning generic representations with tremendous EEG data in BCI. InICLR

    2024

  38. [39]

    Ce Ju and Cuntai Guan. 2023. Graph neural networks on spd manifolds for motor imagery classification: A perspective from the time–frequency analysis.IEEE Trans. Neural Netw. Learn. Syst.(2023), 17701 – 17715

    2023

  39. [40]

    Max Kochurov, Rasul Karimov, and Serge Kozlukov. 2020. Geoopt: Riemannian optimization in pytorch.arXiv preprint arXiv:2005.02819(2020)

    arXiv 2020

  40. [41]

    Vernon J Lawhern, Amelia J Solon, Nicholas R Waytowich, Stephen M Gordon, Chou P Hung, and Brent J Lance. 2018. EEGNet: a compact convolutional neural network for EEG-based brain–computer interfaces.J. Neural Eng.(2018), 056013

    2018

  41. [42]

    Yann LeCun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. 2002. Gradient- based learning applied to document recognition.Proc. IEEE(2002), 2278–2324

    2002

  42. [43]

    Cunbo Li, Tian Tang, Yue Pan, Lei Yang, Shuhan Zhang, Zhaojin Chen, Peiyang Li, Dongrui Gao, Huafu Chen, Fali Li, et al . 2024. An efficient graph learning system for emotion recognition inspired by the cognitive prior graph of EEG brain network.IEEE Trans. Neural Netw. Learn. Syst.(2024), 7130 – 7144

    2024

  43. [44]

    Fabien Lotte, Marco Congedo, Anatole Lécuyer, Fabrice Lamarche, and Bruno Ar- naldi. 2007. A review of classification algorithms for EEG-based brain–computer interfaces.Journal of neural engineering(2007), R1

    2007

  44. [45]

    Luigi Malagò, Luigi Montrucchio, and Giovanni Pistone. 2018. Wasserstein Riemannian geometry of Gaussian densities.Information Geometry(2018), 137– 179

    2018

  45. [46]

    Ravikiran Mane, Effie Chew, Karen Chua, Kai Keng Ang, Neethu Robinson, A Prasad Vinod, Seong-Whan Lee, and Cuntai Guan. 2021. FBCNet: A multi- view convolutional neural network for brain-computer interface.arXiv preprint arXiv:2104.01233(2021)

    arXiv 2021

  46. [47]

    Perrin Margaux, Maby Emmanuel, Daligault Sébastien, Bertrand Olivier, and Mat- tout Jérémie. 2012. Objective and subjective evaluation of online error correction during P300-based spelling.Adv. Hum-Comput. Interact.(2012)

    2012

  47. [48]

    Yazeed K Musallam, Nasser I AlFassam, Ghulam Muhammad, Syed Umar Amin, Mansour Alsulaiman, Wadood Abdul, Hamdi Altaheri, Mohamed A Bencherif, and Mohammed Algabri. 2021. Electroencephalography-based motor imagery classification using temporal convolutional network fusion.Biomed. Signal Process. Control(2021), 102826

    2021

  48. [49]

    Xuan Son Nguyen. 2022. The Gyro-Structure of Some Matrix Manifolds. In NeurIPS

    2022

  49. [50]

    Xuan Son Nguyen, Luc Brun, Olivier Lézoray, and Sébastien Bougleux. 2019. A neural network based on SPD manifold learning for skeleton-based hand gesture recognition. InCVPR

    2019

  50. [51]

    Xuan Son Nguyen and Shuo Yang. 2023. Building neural networks on matrix manifolds: a gyrovector space approach. InICML

    2023

  51. [52]

    Yue-Ting Pan, Jing-Lun Chou, and Chun-Shu Wei. 2022. MAtt: a manifold attention network for EEG decoding. InNeurIPS

    2022

  52. [53]

    Xavier Pennec, Pierre Fillard, and Nicholas Ayache. 2006. A Riemannian frame- work for tensor computing.Int. J. Comput. Vis.(2006), 41–66

    2006

  53. [54]

    Yannick Roy, Hubert Banville, Isabela Albuquerque, Alexandre Gramfort, Tiago H Falk, and Jocelyn Faubert. 2019. Deep learning-based electroencephalography analysis: a systematic review.J. Neural Eng.(2019), 051001

    2019

  54. [55]

    Andres Sanin, Conrad Sanderson, Mehrtash T Harandi, and Brian C Lovell. 2013. Spatio-temporal covariance descriptors for action and gesture recognition. In W ACV

    2013

  55. [56]

    Eduardo Santamaria-Vazquez, Victor Martinez-Cagigal, Fernando Vaquerizo- Villar, and Roberto Hornero. 2020. EEG-inception: a novel deep convolutional neural network for assistive ERP-based brain-computer interfaces.IEEE Trans. Neural Syst. Rehabil. Eng.(2020), 2773–2782

    2020

  56. [57]

    Robin Tibor Schirrmeister, Jost Tobias Springenberg, Lukas Dominique Josef Fiederer, Martin Glasstetter, Katharina Eggensperger, Michael Tangermann, Frank Hutter, Wolfram Burgard, and Tonio Ball. 2017. Deep learning with convolutional neural networks for EEG decoding and visualization.Hum. Brain Mapp.(2017), 5391–5420. Towards Robust EEG Decoding Based on...

    2017

  57. [58]

    Nikolopoulos Spiros. 2016. MAMEM EEG SSVEP Dataset II (256 channels, 11 subjects, 5 frequencies presented simultaneously). (2016)

    2016

  58. [59]

    Yoon-Je Suh and Byung Hyung Kim. 2021. Riemannian embedding banks for common spatial patterns with EEG-based SPD neural networks. InAAAI

    2021

  59. [60]

    Fengzhen Tang, Mengling Fan, and Peter Tiňo. 2020. Generalized learning Riemannian space quantization: A case study on Riemannian manifold of SPD matrices.IEEE Trans. Neural Netw. Learn. Syst.(2020), 281–292

    2020

  60. [61]

    Yann Thanwerdas and Xavier Pennec. 2019. Is affine-invariance well defined on SPD matrices? A principled continuum of metrics. InGeometric Science of Information: 4th International Conference, GSI 2019, Toulouse, France, August 27–29, 2019, Proceedings 4. 502–510

    2019

  61. [62]

    Yann Thanwerdas and Xavier Pennec. 2022. The geometry of mixed-Euclidean metrics on symmetric positive definite matrices.Differ. Geom. Appl.(2022)

    2022

  62. [63]

    Diego Tosato, Michela Farenzena, Mauro Spera, Vittorio Murino, and Marco Cristani. 2010. Multi-class classification on Riemannian manifolds for video surveillance. InECCV

    2010

  63. [64]

    Raviteja Vemulapalli, Jaishanker K Pillai, and Rama Chellappa. 2013. Kernel learning for extrinsic classification of manifold features. InCVPR

    2013

  64. [65]

    Ruiping Wang, Huimin Guo, Larry S Davis, and Qionghai Dai. 2012. Covariance discriminative learning: A natural and efficient approach to image set classifica- tion. InCVPR

    2012

  65. [66]

    Rui Wang, Chen Hu, Ziheng Chen, Xiao-Jun Wu, and Xiaoning Song. 2024. A Grassmannian Manifold Self-Attention Network for Signal Classification. In IJCAI

    2024

  66. [67]

    Rui Wang, Shaocheng Jin, Ziheng Chen, Xiaoqing Luo, and Xiao-Jun Wu. 2025. Learning to Normalize on the SPD Manifold under Bures-Wasserstein Geometry. InCVPR

    2025

  67. [68]

    Rui Wang, Xiao-Jun Wu, and Josef Kittler. 2022. SymNet: A Simple Symmetric Positive Definite Manifold Deep Learning Method for Image Set Classification. IEEE Trans. Neural Netw. Learn. Syst.(2022), 2208–2222

    2022

  68. [69]

    Chun-Shu Wei, Toshiaki Koike-Akino, and Ye Wang. 2019. Spatial component- wise convolutional network (SCCNet) for motor-imagery EEG classification. IEEE/EMBS Conference on Neural Engineering(2019), 328–331

    2019

  69. [70]

    Chun-Shu Wei, Toshiaki Koike-Akino, and Ye Wang. 2019. Spatial Component- wise Convolutional Network (SCCNet) for Motor-Imagery EEG Classification. IEEE/EMBS Conference on Neural Engineering(2019), 328–331

    2019

  70. [71]

    Chaoqi Yang, M Westover, and Jimeng Sun. 2023. Biot: Biosignal transformer for cross-data learning in the wild. InNeurIPS

    2023

  71. [72]

    Le Yang, Marc Arnaudon, and Frédéric Barbaresco. 2010. Riemannian median, geometry of covariance matrices and radar target detection.European Radar Conference(2010), 415–418

    2010

  72. [73]

    Florian Yger. 2013. A review of kernels on covariance matrices for BCI applica- tions. InIEEE international workshop on machine learning for signal processing

    2013

  73. [74]

    Florian Yger, Maxime Berar, and Fabien Lotte. 2016. Riemannian approaches in brain-computer interfaces: a review.IEEE Trans. Neural Syst. Rehabil. Eng.(2016), 1753–1762

    2016

  74. [75]

    Tong Zhang, Wenming Zheng, Zhen Cui, Yuan Zong, Chaolong Li, Xiaoyan Zhou, and Jian Yang. 2020. Deep manifold-to-manifold transforming network for skeleton-based action recognition.IEEE Trans. Multimedia(2020), 2926–2937

    2020

  75. [76]

    Zhijun Zhang, Yu He, Weijian Mai, Yamei Luo, Xiaoli Li, Yuanxiong Cheng, Xiaoming Huang, and Run Lin. 2024. Convolutional Dynamically Convergent Differential Neural Network for Brain Signal Classification.IEEE Trans. Neural Netw. Learn. Syst.(2024), 8166 – 8177

    2024

  76. [77]

    Luping Zhou, Lei Wang, Jianjia Zhang, Yinghuan Shi, and Yang Gao. 2017. Revis- iting metric learning for SPD matrix based visual representation. InCVPR. KDD ’26, August 09–13, 2026, Jeju Island, Republic of Korea Shaocheng Jin et al. A Related Work A.1 Neural Networks for EEG Due to the superior feature extraction ability of deep learning, numerous end-to...

    2017