pith. sign in

arxiv: 2605.29731 · v1 · pith:AFINYY7Rnew · submitted 2026-05-28 · 💻 cs.LG

EMAG: Differentiable 4D Gaussian Mixture Splatting for EEG Spatial Super-Resolution

Alex Lazarovich , Ofir Itzhak Shahar , Gur Elkin , Ohad Ben-Shahar This is my paper

Pith reviewed 2026-06-29 08:43 UTC · model grok-4.3

classification 💻 cs.LG
keywords EEG super-resolutionGaussian mixture model4D Gaussiansdifferentiable renderingbrain source modelingspatial interpolationanisotropic Gaussians
0
0 comments X

The pith

A mixture of anisotropic 4D Gaussians on a spherical grid reconstructs high-density EEG from sparse low-density electrodes via differentiable rendering.

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

The paper presents EMAG to recover dense scalp EEG measurements from far fewer electrodes by modeling cortical sources directly as mixtures of 4D space-time Gaussians. Each Gaussian carries a full 4x4 precision matrix so that spatial spread and temporal coupling can vary by direction and location. Because the mapping from these sources to electrode voltages is written as a differentiable forward model, the system trains end-to-end on observed scalp data alone. A reader would care if the approach works because it removes the need for expensive high-density hardware while still producing interpretable source maps that could support clinical or neuroscientific use.

Core claim

EMAG places multiple anisotropic 4D Gaussians at each point of a spherical brain grid; each Gaussian is parameterized by a full 4x4 precision matrix that encodes anisotropic spatial spreads and explicit space-time coupling. Scalp potentials are obtained by summing the differentiable field contributions of all Gaussians at electrode locations, allowing the entire pipeline to be trained without any explicit source-localization labels. On the Localize-MI, SEED, and SEED-IV benchmarks the resulting reconstructions exceed prior super-resolution methods at most factors between 2x and 16x.

What carries the argument

Mixture of anisotropic 4D space-time Gaussians, each defined by a 4x4 precision matrix placed on a spherical brain grid, whose differentiable summation produces electrode voltages.

If this is right

  • Outperforms the prior state-of-the-art EEG super-resolution method at most factors on Localize-MI, SEED, and SEED-IV.
  • Yields direct visualizations of the learned source configurations without additional post-processing.
  • Supports downstream tasks such as source localization and biomarker discovery from the explicit Gaussian parameters.
  • Trains without any supervised source-localization targets.

Where Pith is reading between the lines

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

  • The same 4D Gaussian splatting could be tested on simultaneous EEG-fMRI recordings to check whether the recovered sources align with BOLD activation maps.
  • Because the representation separates spatial and temporal axes, it may allow controlled experiments on how temporal smoothing affects spatial resolution in real-time BCI settings.
  • The spherical-grid placement could be replaced by a data-driven point set to test whether performance gains come from the Gaussian form or from the grid geometry.

Load-bearing premise

Brain electrical sources can be accurately represented as a mixture of anisotropic 4D space-time Gaussians placed on a spherical grid.

What would settle it

EMAG would be falsified if it produced lower reconstruction accuracy than simple interpolation baselines on a new EEG recording set whose electrode layout or head geometry differs markedly from the training benchmarks.

Figures

Figures reproduced from arXiv: 2605.29731 by Alex Lazarovich, Gur Elkin, Ofir Itzhak Shahar, Ohad Ben-Shahar.

Figure 3
Figure 3. Figure 3: Subject [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Recording geometries used in this paper. SEED/SEED-IV (left, middle) use the standard [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Named electrode subsets on the 62-channel SEED/SEED-IV cap (top-down view, anterior [PITH_FULL_IMAGE:figures/full_fig_p018_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Top-10 learned 4D Gaussians rendered as 1σ spatial ellipsoids (brightness ∝ |wn|) versus ground-truth stimulation midpoints (cyan ⋆), in T1w-native space inside the brain envelope (r=90 mm). Subjects shown: best (sub-01), median (sub-05), and worst (sub-06). F.2 Anisotropy distribution [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Reconstructed scalp amplitude in the post-stim window (mean [PITH_FULL_IMAGE:figures/full_fig_p025_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Learned Gaussian amplitudes (lateral and axial views) for SEED sub-01 (top) and Localize [PITH_FULL_IMAGE:figures/full_fig_p025_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Anisotropy of the trained 4D Gaussian field (top: SEED sub-01, bottom: Localize-MI [PITH_FULL_IMAGE:figures/full_fig_p026_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Localize-MI sub-01, channel 100, trial 0: ground-truth high-density signal (black) vs. [PITH_FULL_IMAGE:figures/full_fig_p026_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Welch PSD: true (black) vs. EMAG (crimson). Top: Localize-MI sub-01 ( [PITH_FULL_IMAGE:figures/full_fig_p027_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Single-time-step electrode-space snapshots (top: Localize-MI sub-01, bottom: SEED sub [PITH_FULL_IMAGE:figures/full_fig_p028_12.png] view at source ↗
read the original abstract

High-density electroencephalography (HD-EEG) enables fine-grained measurement of cortical activity but requires expensive hardware and lengthy setup times, limiting its clinical and research accessibility. We propose EMAG (EEG Mixture of Anisotropic Gaussians), a differentiable framework that reconstructs HD-EEG signals from a sparse subset of low-density (LD) electrodes by representing brain electrical sources as a mixture of anisotropic 4D space-time Gaussians. EMAG places a mixture of multiple Gaussians at each point of a spherical brain grid, each parameterized by a full 4 x 4 precision matrix, enabling anisotropic spatial spreads and explicit coupling between spatial and temporal dimensions. The forward model renders scalp EEG via differentiable Gaussian field contributions at electrode locations, enabling end-to-end training without explicit source localization supervision. We evaluate EMAG on three public EEG benchmarks (Localize-MI, SEED, and SEED-IV) at super-resolution factors of 2x through 8/16x. EMAG outperforms the current state-of-the-art EEG super-resolution method at most super-resolution factors on three standard benchmarks (Localize-MI, SEED, SEED-IV). The explicit Gaussian parameterization further enables direct visualization and interpretability of learned brain source configurations, potentially opening avenues for clinical and neuroscientific applications, such as source localization or biomarker discovery.

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

2 major / 0 minor

Summary. The paper proposes EMAG, a differentiable framework representing brain electrical sources as mixtures of anisotropic 4D space-time Gaussians placed at points on a spherical brain grid. Each Gaussian uses a full 4x4 precision matrix to capture anisotropic spatial spreads and space-time coupling. Scalp EEG is reconstructed from sparse low-density electrodes via differentiable Gaussian field contributions at electrode locations, enabling end-to-end training without explicit source localization supervision. The method is evaluated on Localize-MI, SEED, and SEED-IV at 2x to 8/16x super-resolution factors and claims to outperform the current SOTA EEG super-resolution method at most factors.

Significance. If the Gaussian mixture model and differentiable rendering accurately capture the underlying source dynamics and forward propagation, the work could advance accessible HD-EEG reconstruction while providing interpretable visualizations of learned source configurations. The explicit parameterization and lack of need for source supervision are potential strengths for neuroscientific applications. However, the significance is tempered by the absence of biophysical grounding in the provided description.

major comments (2)
  1. [Abstract] Abstract: the central modeling choice replaces the standard quasi-static Maxwell forward problem (lead-field matrix from realistic head geometry and conductivity) with differentiable Gaussian field contributions on a spherical grid. This is load-bearing for the reconstruction claim, yet no validation against biophysical simulators is described, leaving open the possibility that benchmark gains arise from data fitting rather than faithful volume conduction modeling.
  2. [Abstract] Abstract: the weakest assumption—that mixtures of anisotropic 4D Gaussians on a spherical grid suffice for accurate scalp potential reconstruction without explicit supervision—is not accompanied by any reported comparison to physics-based forward models or ablation on grid resolution/precision matrix constraints, undermining the interpretability and generalization claims at high super-resolution factors.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the biophysical aspects of EMAG. We address the two major comments point by point below and commit to revisions that strengthen the manuscript's claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central modeling choice replaces the standard quasi-static Maxwell forward problem (lead-field matrix from realistic head geometry and conductivity) with differentiable Gaussian field contributions on a spherical grid. This is load-bearing for the reconstruction claim, yet no validation against biophysical simulators is described, leaving open the possibility that benchmark gains arise from data fitting rather than faithful volume conduction modeling.

    Authors: EMAG is explicitly designed as a data-driven, end-to-end differentiable surrogate that learns effective source representations and scalp mappings directly from observed EEG without requiring explicit source localization or biophysical forward modeling. This choice prioritizes accessibility and training on real datasets over explicit Maxwell-equation compliance. We agree that the absence of validation against biophysical simulators (e.g., lead-field matrices from realistic head models) leaves open the possibility that gains are due to data fitting. In the revised manuscript we will add such validation experiments on simulated data generated with standard forward models to quantify how well the learned Gaussian contributions approximate volume conduction. revision: yes

  2. Referee: [Abstract] Abstract: the weakest assumption—that mixtures of anisotropic 4D Gaussians on a spherical grid suffice for accurate scalp potential reconstruction without explicit supervision—is not accompanied by any reported comparison to physics-based forward models or ablation on grid resolution/precision matrix constraints, undermining the interpretability and generalization claims at high super-resolution factors.

    Authors: The empirical results on three benchmarks at 2x–16x factors provide evidence that the Gaussian-mixture representation suffices for the super-resolution task. We acknowledge that direct comparisons to physics-based forward models and ablations on grid resolution and precision-matrix constraints are missing. The revised manuscript will incorporate both: (i) side-by-side reconstruction comparisons against lead-field-based baselines on simulated data, and (ii) ablations varying grid density and precision-matrix constraints to support the interpretability and generalization statements. revision: yes

Circularity Check

0 steps flagged

No circularity; modeling choice and empirical benchmarks are independent

full rationale

The paper introduces a 4D Gaussian mixture representation and differentiable rendering as a modeling choice for EEG super-resolution, trained end-to-end on sparse-to-dense reconstruction without explicit localization labels. No quoted equations or steps reduce a claimed prediction to a fitted input by construction, nor rely on self-citation chains, uniqueness theorems, or renamed known results. Performance claims are benchmark comparisons (Localize-MI, SEED, SEED-IV) external to the model definition itself. The derivation is self-contained against standard supervised learning practices.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Only abstract available, so ledger is necessarily incomplete. The central claim rests on the modeling choice of 4D Gaussians and the assumption that differentiable rendering suffices without source supervision.

invented entities (1)
  • Mixture of anisotropic 4D space-time Gaussians on spherical brain grid no independent evidence
    purpose: Represent brain electrical sources for differentiable forward modeling of scalp EEG
    Core representation stated in the abstract; no independent evidence supplied.

pith-pipeline@v0.9.1-grok · 5782 in / 1158 out tokens · 27861 ms · 2026-06-29T08:43:45.878328+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

45 extracted references · 5 canonical work pages · 1 internal anchor

  1. [1]

    Eeg channel selection techniques in motor imagery applications: A review and new perspectives.Bioengineering, 9(12):726, 2022

    Abdullah, Ibrahima Faye, and Md Rafiqul Islam. Eeg channel selection techniques in motor imagery applications: A review and new perspectives.Bioengineering, 9(12):726, 2022

    2022

  2. [2]

    Deep eeg super-resolution: Upsampling eeg spatial resolu- tion with generative adversarial networks

    Isaac A Corley and Yufei Huang. Deep eeg super-resolution: Upsampling eeg spatial resolu- tion with generative adversarial networks. In2018 IEEE EMBS international conference on biomedical & health informatics (BHI), pages 100–103. IEEE, 2018

    2018

  3. [3]

    Dynamic statistical parametric mapping: combining fmri and meg for high-resolution imaging of cortical activity.neuron, 26(1):55–67, 2000

    Anders M Dale, Arthur K Liu, Bruce R Fischl, Randy L Buckner, John W Belliveau, Jeffrey D Lewine, and Eric Halgren. Dynamic statistical parametric mapping: combining fmri and meg for high-resolution imaging of cortical activity.neuron, 26(1):55–67, 2000

    2000

  4. [4]

    Anterior cerebral asymmetry and the nature of emotion.Brain and cognition, 20(1):125–151, 1992

    Richard J Davidson. Anterior cerebral asymmetry and the nature of emotion.Brain and cognition, 20(1):125–151, 1992

    1992

  5. [5]

    Differential entropy feature for eeg-based emotion classification

    Ruo-Nan Duan, Jia-Yi Zhu, and Bao-Liang Lu. Differential entropy feature for eeg-based emotion classification. In2013 6th international IEEE/EMBS conference on neural engineering (NER), pages 81–84. IEEE, 2013

    2013

  6. [6]

    4d-rotor gaussian splatting: towards efficient novel view synthesis for dynamic scenes

    Yuanxing Duan, Fangyin Wei, Qiyu Dai, Yuhang He, Wenzheng Chen, and Baoquan Chen. 4d-rotor gaussian splatting: towards efficient novel view synthesis for dynamic scenes. InACM SIGGRAPH 2024 Conference Papers, pages 1–11, 2024

    2024

  7. [7]

    Mne software for processing meg and eeg data.neuroimage, 86:446–460, 2014

    Alexandre Gramfort, Martin Luessi, Eric Larson, Denis A Engemann, Daniel Strohmeier, Christian Brodbeck, Lauri Parkkonen, and Matti S Hämäläinen. Mne software for processing meg and eeg data.neuroimage, 86:446–460, 2014

    2014

  8. [8]

    Review on solving the inverse problem in eeg source analysis.Journal of neuroengineering and rehabilitation, 5(1): 25, 2008

    Roberta Grech, Tracey Cassar, Joseph Muscat, Kenneth P Camilleri, Simon G Fabri, Michalis Zervakis, Petros Xanthopoulos, Vangelis Sakkalis, and Bart Vanrumste. Review on solving the inverse problem in eeg source analysis.Journal of neuroengineering and rehabilitation, 5(1): 25, 2008

    2008

  9. [9]

    Review on solving the forward problem in eeg source analysis.Journal of neuroengineering and rehabilitation, 4(1):46, 2007

    Hans Hallez, Bart Vanrumste, Roberta Grech, Joseph Muscat, Wim De Clercq, Anneleen Vergult, Yves D’Asseler, Kenneth P Camilleri, Simon G Fabri, Sabine Van Huffel, et al. Review on solving the forward problem in eeg source analysis.Journal of neuroengineering and rehabilitation, 4(1):46, 2007

    2007

  10. [10]

    Interpreting magnetic fields of the brain: minimum norm estimates.Medical & biological engineering & computing, 32(1):35–42, 1994

    Matti S Hämäläinen and Risto J Ilmoniemi. Interpreting magnetic fields of the brain: minimum norm estimates.Medical & biological engineering & computing, 32(1):35–42, 1994

    1994

  11. [11]

    Electromagnetic source imaging via a data-synthesis-based convolutional encoder–decoder network.IEEE Transactions on Neural Networks and Learning Systems, 35(5):6423–6437, 2022

    Gexin Huang, Ke Liu, Jiawen Liang, Chang Cai, Zheng Hui Gu, Feifei Qi, Yuanqing Li, Zhu Liang Yu, and Wei Wu. Electromagnetic source imaging via a data-synthesis-based convolutional encoder–decoder network.IEEE Transactions on Neural Networks and Learning Systems, 35(5):6423–6437, 2022

    2022

  12. [12]

    Self-attention-based diffusion model for time-series imputation in partial blackout scenarios

    Mohammad Rafid Ul Islam, Prasad Tadepalli, and Alan Fern. Self-attention-based diffusion model for time-series imputation in partial blackout scenarios. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 17564–17572, 2025

    2025

  13. [13]

    Multi-modal electrophys- iological source imaging with attention neural networks based on deep fusion of eeg and meg

    Meng Jiao, Shihao Yang, Xiaochen Xian, Neel Fotedar, and Feng Liu. Multi-modal electrophys- iological source imaging with attention neural networks based on deep fusion of eeg and meg. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 32:2492–2502, 2024

    2024

  14. [14]

    Neural brain fields: A nerf-inspired approach for generating nonexistent eeg electrodes.arXiv preprint arXiv:2601.00012, 2025

    Shahar Ain Kedem, Itamar Zimerman, and Eliya Nachmani. Neural brain fields: A nerf-inspired approach for generating nonexistent eeg electrodes.arXiv preprint arXiv:2601.00012, 2025

    work page arXiv 2025

  15. [15]

    3d gaussian splatting for real-time radiance field rendering.ACM Trans

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, George Drettakis, et al. 3d gaussian splatting for real-time radiance field rendering.ACM Trans. Graph., 42(4):139–1, 2023

    2023

  16. [16]

    Miniaturization for wearable eeg systems: recording hardware and data processing.Biomedical Engineering Letters, 12(3):239–250, 2022

    Minjae Kim, Seungjae Yoo, and Chul Kim. Miniaturization for wearable eeg systems: recording hardware and data processing.Biomedical Engineering Letters, 12(3):239–250, 2022

    2022

  17. [17]

    Adam: A Method for Stochastic Optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014. 10

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  18. [18]

    Eegnet: a compact convolutional neural network for eeg-based brain–computer interfaces.Journal of neural engineering, 15(5):056013, 2018

    Vernon J Lawhern, Amelia J Solon, Nicholas R Waytowich, Stephen M Gordon, Chou P Hung, and Brent J Lance. Eegnet: a compact convolutional neural network for eeg-based brain–computer interfaces.Journal of neural engineering, 15(5):056013, 2018

    2018

  19. [19]

    Estformer: Transformer utilizing spatiotemporal dependencies for eeg super-resolution.arXiv e-prints, pages arXiv–2312, 2023

    Dongdong Li, Zhongliang Zeng, Zhe Wang, and Hai Yang. Estformer: Transformer utilizing spatiotemporal dependencies for eeg super-resolution.arXiv e-prints, pages arXiv–2312, 2023

    2023

  20. [20]

    Step-aware residual-guided diffusion for eeg spatial super-resolution.arXiv preprint arXiv:2510.19166, 2025

    Hongjun Liu, Leyu Zhou, Zijianghao Yang, and Chao Yao. Step-aware residual-guided diffusion for eeg spatial super-resolution.arXiv preprint arXiv:2510.19166, 2025

    work page arXiv 2025

  21. [21]

    Assigning channel weights using an attention mechanism: an eeg interpolation algorithm.Frontiers in neuroscience, 17:1251677, 2023

    Renjie Liu, Zaijun Wang, Jiang Qiu, and Xue Wang. Assigning channel weights using an attention mechanism: an eeg interpolation algorithm.Frontiers in neuroscience, 17:1251677, 2023

    2023

  22. [22]

    Rdpi: a refine diffusion probability generation method for spatiotemporal data imputation

    Zijin Liu, Xiang Zhao, and You Song. Rdpi: a refine diffusion probability generation method for spatiotemporal data imputation. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 12255–12263, 2025

    2025

  23. [23]

    Eeg source imaging: a practical review of the analysis steps.Frontiers in neurology, 10:325, 2019

    Christoph M Michel and Denis Brunet. Eeg source imaging: a practical review of the analysis steps.Frontiers in neurology, 10:325, 2019

    2019

  24. [24]

    Simultaneous human intracerebral stimulation and hd-eeg, ground-truth for source localization methods

    Ezequiel Mikulan, Simone Russo, Sara Parmigiani, Simone Sarasso, Flavia Maria Zauli, Annal- isa Rubino, Pietro Avanzini, Anna Cattani, Alberto Sorrentino, Steve Gibbs, et al. Simultaneous human intracerebral stimulation and hd-eeg, ground-truth for source localization methods. Scientific data, 7(1):127, 2020

    2020

  25. [25]

    Nerf: Representing scenes as neural radiance fields for view synthesis

    Ben Mildenhall, Pratul P Srinivasan, Matthew Tancik, Jonathan T Barron, Ravi Ramamoor- thi, and Ren Ng. Nerf: Representing scenes as neural radiance fields for view synthesis. Communications of the ACM, 65(1):99–106, 2021

    2021

  26. [26]

    Standardized low-resolution brain electromagnetic tomography (sloreta): technical details.Methods find exp clin pharmacol, 24(Suppl D):5–12, 2002

    Roberto Domingo Pascual-Marqui et al. Standardized low-resolution brain electromagnetic tomography (sloreta): technical details.Methods find exp clin pharmacol, 24(Suppl D):5–12, 2002

    2002

  27. [27]

    Eeg channel reconstruction using con- volutional neural networks in limited bcis: A proposed method for neuromarketing applications

    Mario Quiles Pérez, Sergio López Bernal, Eduardo Horna Prat, Luis Montesano Del Campo, Lorenzo Fernández Maimó, and Alberto Huertas Celdrán. Eeg channel reconstruction using con- volutional neural networks in limited bcis: A proposed method for neuromarketing applications. Applied Soft Computing, 181:113455, 2025

    2025

  28. [28]

    Spherical splines for scalp potential and current density mapping.Electroencephalography and clinical neurophysiology, 72(2):184–187, 1989

    François Perrin, Jacques Pernier, Olivier Bertrand, and Jean Francois Echallier. Spherical splines for scalp potential and current density mapping.Electroencephalography and clinical neurophysiology, 72(2):184–187, 1989

    1989

  29. [29]

    Brain computer interface: control signals review.Neurocomputing, 223:26–44, 2017

    Rabie A Ramadan and Athanasios V Vasilakos. Brain computer interface: control signals review.Neurocomputing, 223:26–44, 2017

    2017

  30. [30]

    Eeg channel interpolation using deep encoder-decoder networks

    Sari Saba-Sadiya, Tuka Alhanai, Taosheng Liu, and Mohammad M Ghassemi. Eeg channel interpolation using deep encoder-decoder networks. In2020 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pages 2432–2439. IEEE, 2020

    2020

  31. [31]

    Rui Sun, Abbas Sohrabpour, Gregory A Worrell, and Bin He. Deep neural networks constrained by neural mass models improve electrophysiological source imaging of spatiotemporal brain dynamics.Proceedings of the National Academy of Sciences, 119(31):e2201128119, 2022

    2022

  32. [32]

    Fourier features let networks learn high frequency functions in low dimensional domains.Advances in neural information processing systems, 33:7537–7547, 2020

    Matthew Tancik, Pratul Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan Barron, and Ren Ng. Fourier features let networks learn high frequency functions in low dimensional domains.Advances in neural information processing systems, 33:7537–7547, 2020

    2020

  33. [33]

    Identifying relevant eeg channels for subject- independent emotion recognition using attention network layers.Frontiers in psychiatry, 16: 1494369, 2025

    Camilo E Valderrama and Anshul Sheoran. Identifying relevant eeg channels for subject- independent emotion recognition using attention network layers.Frontiers in psychiatry, 16: 1494369, 2025. 11

    2025

  34. [34]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

    2017

  35. [35]

    Generating realistic neurophysiological time series with denoising diffusion probabilistic models.Patterns, 5(9), 2024

    Julius Vetter, Jakob H Macke, and Richard Gao. Generating realistic neurophysiological time series with denoising diffusion probabilistic models.Patterns, 5(9), 2024

    2024

  36. [36]

    Generative ai enables eeg super-resolution via spatio-temporal adaptive diffusion learning.IEEE Transactions on Consumer Electronics, 71(1):1034–1045, 2025

    Shuqiang Wang, Tong Zhou, Yanyan Shen, Ye Li, Guoheng Huang, and Yong Hu. Generative ai enables eeg super-resolution via spatio-temporal adaptive diffusion learning.IEEE Transactions on Consumer Electronics, 71(1):1034–1045, 2025

    2025

  37. [37]

    Brain-computer interfaces (bcis) for communication and control

    Jonathan R Wolpaw. Brain-computer interfaces (bcis) for communication and control. InPro- ceedings of the 9th international ACM SIGACCESS conference on Computers and accessibility, pages 1–2, 2007

    2007

  38. [38]

    4d gaussian splatting for real-time dynamic scene rendering

    Guanjun Wu, Taoran Yi, Jiemin Fang, Lingxi Xie, Xiaopeng Zhang, Wei Wei, Wenyu Liu, Qi Tian, and Xinggang Wang. 4d gaussian splatting for real-time dynamic scene rendering. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 20310–20320, 2024

    2024

  39. [39]

    Sasdim: Self-adaptive noise scaling diffusion model for spatial time series imputation.arXiv preprint arXiv:2309.01988, 2023

    Shunyang Zhang, Senzhang Wang, Xianzhen Tan, Ruochen Liu, Jian Zhang, and Jianxin Wang. Sasdim: Self-adaptive noise scaling diffusion model for spatial time series imputation.arXiv preprint arXiv:2309.01988, 2023

    work page arXiv 2023

  40. [40]

    Wei-Long Zheng and Bao-Liang Lu. Investigating critical frequency bands and channels for eeg-based emotion recognition with deep neural networks.IEEE Transactions on autonomous mental development, 7(3):162–175, 2015

    2015

  41. [41]

    Emotionmeter: A multimodal framework for recognizing human emotions.IEEE transactions on cybernetics, 49 (3):1110–1122, 2018

    Wei-Long Zheng, Wei Liu, Yifei Lu, Bao-Liang Lu, and Andrzej Cichocki. Emotionmeter: A multimodal framework for recognizing human emotions.IEEE transactions on cybernetics, 49 (3):1110–1122, 2018. A Implementation details Brain grid.The brain domain spans [−90,90]mm in each spatial dimension. The brain grid at resolution R consists of all points on a unif...

    2018

  42. [42]

    loading the EMAG model trained on subjecti(split seeds)

  43. [43]

    feeding it subject j’s LD test recordings (the same LD subset and test split that was used when evaluating subject j in the main results, so T (s) jj exactly reproduces the within-subject number)

  44. [44]

    circumference

    recording NMSE, PCC and SNR between the model’s output and subject j’s HD ground truth. We report within(s) = 1 S X i T (s) ii ,cross (s) = 1 S(S−1) X i̸=j T (s) ij , and average across seeds. Each non-diagonal cell T (s) ij uses no information from subject j during training: subject i’s 4D Gaussian field is applied verbatim to subjectj’s LD signal, with ...

    work page arXiv 2023

  45. [45]

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

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