pith. sign in

arxiv: 2605.10337 · v1 · submitted 2026-05-11 · 💻 cs.AI · eess.SP

CORTEG: Foundation Models Enable Cross-Modality Representation Transfer from Scalp to Intracranial Brain Recordings

Liuyin Yang , Qiang Sun , Bob Van Dyck , Eva Calvo Merino , Marc M. Van Hulle This is my paper

Pith reviewed 2026-05-12 04:43 UTC · model grok-4.3

classification 💻 cs.AI eess.SP
keywords cross-modality transferEEG foundation modelsECoG decodingbrain-computer interfacesrepresentation transferneural decodingfinger trajectory regressionaudio envelope regression
0
0 comments X

The pith

Scalp EEG foundation models can be adapted to decode intracranial ECoG signals competitively with minimal new data.

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

The paper investigates whether large pretrained models from scalp EEG recordings can be transferred to intracranial ECoG data for brain-computer interface tasks. It proposes a framework that leverages these models to enable cross-patient learning and fast calibration for new patients. This matters because ECoG provides high-quality signals but collecting enough data per patient is challenging, so successful transfer could make advanced BCIs more accessible. The approach is tested on finger movement and audio decoding tasks where it performs at or above specialized methods.

Core claim

CORTEG demonstrates that representations learned from scalp-EEG foundation models contain transferable information to ECoG, allowing a pretrained backbone combined with a spatial adapter and dual-stream tokenizer to match or exceed task-specific baselines on finger trajectory and audio envelope regression, with notable advantages in low-data calibration scenarios for new patients.

What carries the argument

The CORTEG framework, which integrates a pretrained EEG foundation model backbone, an electrode-aware KNNSoftFourier spatial adapter, a dual-stream tokenizer for low-frequency and high-gamma activity, and leave-one-subject-out fine-tuning.

If this is right

  • CORTEG enables competitive decoding performance on ECoG without relying solely on subject-specific training.
  • It supports cross-patient learning by adapting from large scalp datasets to individual intracranial recordings.
  • Per-patient calibration can be completed in 10-30 minutes on a single GPU for practical use.
  • Feature analyses from the model align with established neurophysiological patterns.
  • Latent representations capture low-dimensional structures in movements like finger trajectories.

Where Pith is reading between the lines

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

  • This suggests that non-invasive scalp data can serve as a foundation for improving invasive neural interfaces in data-limited settings.
  • Similar transfer strategies might apply to other brain signal modalities or decoding tasks beyond regression.
  • Future work could test the framework on additional ECoG datasets to confirm generalizability across different electrode configurations.

Load-bearing premise

The representations learned from scalp EEG contain information that remains useful for ECoG despite variations in signal noise, electrode density, and the specific brain areas covered.

What would settle it

Demonstrating no performance gain or a performance drop when applying the transferred model to a new ECoG dataset compared to training a model from scratch on that same limited data.

Figures

Figures reproduced from arXiv: 2605.10337 by Bob Van Dyck, Eva Calvo Merino, Liuyin Yang, Marc M. Van Hulle, Qiang Sun.

Figure 1
Figure 1. Figure 1: Overview of CORTEG for cross-modality transfer from scalp EEG to ECoG. (a) A masked autoencoder is pretrained on 128 Hz scalp EEG, learning temporal representations and an EEG channel-embedding codebook E ∈ RM′×D. (b) The KNNSoftFourier spatial adapter maps subject-specific ECoG electrode coordinates q = (x, y, z) into the pretrained EEG channel￾embedding space using a soft codebook-interpolation branch an… view at source ↗
Figure 2
Figure 2. Figure 2: CORTEG decoding results. (a) Full-data method comparison (white = finger, n=9; grey = audio, n=16); error bars ± std; stars: paired Wilcoxon vs. CORTEG-pooled (Bonferroni corrected). (b) Backbone-size scaling: half-violin/box per-subject r at Small / Base / Large CORTEG; best validation model for each. (c) Compute–performance trade-off: training vs. inference time per backbone (marker area ∝ total params; … view at source ↗
Figure 3
Figure 3. Figure 3: Electrode importance and neural manifold of CORTEG. (a, b) Per-electrode importance on a template cortex (top view) for the finger trajectory (a) and audio envelope (b) tasks. The two tasks yield anatomically distinct cortex (sensorimotor vs. superior-temporal) under the same architecture and adapter. (c, d) 3D PCA of CORTEG’s per-window latent vector z (the mean of all transformer￾output tokens taken just… view at source ↗
Figure 4
Figure 4. Figure 4: Finger task: per-subject electrode implantations on MNI fsaverage (left- or right-lateral [PITH_FULL_IMAGE:figures/full_fig_p020_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Audio task: per-subject electrode implantations on MNI fsaverage. Panel S [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Soft-branch attention w(q) for a representative ECoG electrode (subject bp, MNI (−32, +39, +44) mm; left frontal/precentral). Left: the subject’s full ECoG implantation on fsav￾erage, with the highlighted electrode in red. Center: attention at initialization, equal to the k-NN Gaussian kernel over EEG positions (k=8, σ= median nearest-neighbour distance) — the Nadaraya– Watson estimator. Right: attention a… view at source ↗
Figure 7
Figure 7. Figure 7: Per-channel reliability-weighted importance [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of three projection methods for the neural manifold on a representative finger [PITH_FULL_IMAGE:figures/full_fig_p024_8.png] view at source ↗
read the original abstract

Intracranial electrocorticography (ECoG) offers high-signal-to-noise access to cortical activity for brain-computer interfaces, yet limited per-patient data has led most prior work to rely on small, subject-specific decoders that neglect information shared across patients. We investigate whether large pretrained scalp-EEG foundation models (EEG FMs) can be adapted to ECoG, enabling cross-patient learning and competitive decoding performance while calibrating to a held-out patient in 10-30 minutes on a single GPU. We introduce CORTEG, a cross-modality transfer framework that combines a pretrained EEG FM backbone, an electrode-aware KNNSoftFourier spatial adapter, a dual-stream tokenizer for low-frequency and high-gamma activity, and a leave-one-subject-out fine-tuning strategy. We evaluate CORTEG on two challenging regression tasks: public finger trajectory regression (n=9) and private audio envelope regression (n=16). CORTEG matches or exceeds the strongest task-specific baselines on both tasks: it reaches the highest mean correlation among compared methods on the public finger benchmark (gain not statistically significant on n=9 subjects), with larger and statistically significant gains on the audio task and in low-data per-patient calibration. Feature analyses align with neurophysiology, and latent manifolds capture low-dimensional finger-movement structure. CORTEG provides systematic evidence that scalp-EEG pretraining can be repurposed for ECoG decoding, enabling data-efficient intracranial BCIs that can adapt to new patients.

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 / 3 minor

Summary. The manuscript introduces CORTEG, a framework for adapting pretrained scalp-EEG foundation models to intracranial ECoG recordings. It combines a pretrained EEG FM backbone with an electrode-aware KNNSoftFourier spatial adapter, a dual-stream tokenizer separating low-frequency and high-gamma activity, and leave-one-subject-out fine-tuning. The approach is evaluated on finger trajectory regression using a public dataset (n=9 subjects) and audio envelope regression using a private dataset (n=16 subjects), with claims that it matches or exceeds task-specific baselines, achieving the highest mean correlation on the finger task (not statistically significant) and larger, statistically significant gains on the audio task and in low-data per-patient calibration scenarios. Feature analyses are reported to align with neurophysiology and latent spaces capture movement structure.

Significance. If the central claims hold after addressing the noted gaps, the work would offer systematic evidence that large-scale scalp-EEG pretraining can be repurposed for ECoG decoding. This could advance data-efficient intracranial BCIs by reducing per-patient calibration time to 10-30 minutes and enabling cross-patient learning, with potential implications for clinical BCI applications where data scarcity is a limiting factor.

major comments (2)
  1. [Methods and Results] The load-bearing claim that pretrained scalp-EEG foundation model representations drive cross-modality transfer to ECoG is not isolated from the effects of the adapters and training procedure. No ablation is presented that freezes or randomizes the backbone weights while retaining the electrode-aware KNNSoftFourier adapter, dual-stream tokenizer, and leave-one-subject-out fine-tuning; without this control, it remains unclear whether the reported gains (particularly the statistically significant improvements on audio regression and low-data calibration) arise from transferable representations learned on scalp data or from the adapter architecture and fine-tuning recipe alone.
  2. [Abstract and Results] On the public finger trajectory benchmark (n=9 subjects), the highest mean correlation is reported as not statistically significant relative to the strongest baselines. Given the small sample size, this weakens support for the broader claim of matching or exceeding task-specific methods across both tasks, especially since the audio task (n=16, private data) carries the weight of the statistically significant results.
minor comments (3)
  1. Exact implementation details for the compared baselines, full numerical metrics with error bars or confidence intervals, and subject/trial exclusion criteria are not fully specified, which hinders precise reproduction and assessment of the performance claims.
  2. The private audio dataset (n=16) precludes independent verification of the key statistically significant gains; public release of the data or additional public benchmarks would strengthen the work.
  3. Notation for the dual-stream tokenizer and KNNSoftFourier adapter could be clarified with explicit equations or pseudocode to improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thorough and constructive comments on our manuscript. We address each major comment below and outline the revisions we will make to strengthen the paper.

read point-by-point responses

  1. Referee: [Methods and Results] The load-bearing claim that pretrained scalp-EEG foundation model representations drive cross-modality transfer to ECoG is not isolated from the effects of the adapters and training procedure. No ablation is presented that freezes or randomizes the backbone weights while retaining the electrode-aware KNNSoftFourier adapter, dual-stream tokenizer, and leave-one-subject-out fine-tuning; without this control, it remains unclear whether the reported gains (particularly the statistically significant improvements on audio regression and low-data calibration) arise from transferable representations learned on scalp data or from the adapter architecture and fine-tuning recipe alone.

    Authors: We agree that isolating the contribution of the pretrained backbone is crucial for validating our central claim. In the revised manuscript, we will include an ablation study where the backbone weights are frozen during fine-tuning, as well as a control with randomized backbone weights, while retaining the adapter, tokenizer, and fine-tuning procedure. This will help demonstrate that the performance improvements are driven by the transferable representations from the scalp-EEG foundation model rather than the adapter architecture alone. revision: yes

  2. Referee: [Abstract and Results] On the public finger trajectory benchmark (n=9 subjects), the highest mean correlation is reported as not statistically significant relative to the strongest baselines. Given the small sample size, this weakens support for the broader claim of matching or exceeding task-specific methods across both tasks, especially since the audio task (n=16, private data) carries the weight of the statistically significant results.

    Authors: We acknowledge the limitation posed by the small sample size (n=9) on the finger trajectory task, where the improvement is not statistically significant. This is already noted in the manuscript. To address this, we will revise the abstract and results section to more carefully qualify the claims: CORTEG matches the strongest baselines on the finger task and achieves statistically significant improvements on the audio task and in low-data per-patient scenarios. We will also emphasize the consistency across tasks and the practical benefits in data-efficient settings. revision: partial

Circularity Check

0 steps flagged

No circularity: purely empirical framework evaluation

full rationale

The paper presents CORTEG as an empirical transfer framework evaluated via direct comparisons to task-specific baselines on two regression tasks. No mathematical derivations, equations, predictions, or uniqueness theorems are invoked. Performance metrics arise from standard training and testing procedures rather than any self-referential fitting or renaming. Self-citations (if present for the pretrained backbone) are not load-bearing for any claimed derivation, as the central results are falsifiable experimental outcomes. This is self-contained empirical work with no reduction of outputs to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

Abstract supplies insufficient detail to enumerate all free parameters or background axioms; the ledger therefore records only the explicitly introduced components and the core transfer assumption.

axioms (1)
  • domain assumption Representations learned by scalp-EEG foundation models contain information transferable to intracranial ECoG signals
    This premise underpins the entire adaptation framework and cross-modality claim.
invented entities (2)
  • electrode-aware KNNSoftFourier spatial adapter no independent evidence
    purpose: Adapt spatial electrode information from scalp to intracranial placements
    New component introduced to handle modality-specific geometry
  • dual-stream tokenizer for low-frequency and high-gamma activity no independent evidence
    purpose: Separate processing of distinct frequency bands in ECoG signals
    New tokenizer design within the CORTEG pipeline

pith-pipeline@v0.9.0 · 5590 in / 1485 out tokens · 53218 ms · 2026-05-12T04:43:34.765270+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

43 extracted references · 43 canonical work pages · 1 internal anchor

  1. [1]

    An exoskeleton controlled by an epidural wireless brain–machine interface in a tetraplegic patient: a proof-of-concept demonstration.The Lancet Neurology, 18(12):1112–1122, 2019

    Alim Louis Benabid, Thomas Costecalde, Andrey Eliseyev, Guillaume Charvet, Alexandre Verney, Serpil Karakas, Michael Foerster, Aurélien Lambert, Boris Morinière, Neil Abroug, et al. An exoskeleton controlled by an epidural wireless brain–machine interface in a tetraplegic patient: a proof-of-concept demonstration.The Lancet Neurology, 18(12):1112–1122, 2019

    work page 2019

  2. [2]

    Walking naturally after spinal cord injury using a brain–spine interface.Nature, 618(7963):126–133, 2023

    Henri Lorach, Andrea Galvez, Valeria Spagnolo, Felix Martel, Serpil Karakas, Nadine Intering, Molywan Vat, Olivier Faivre, Cathal Harte, Salif Komi, et al. Walking naturally after spinal cord injury using a brain–spine interface.Nature, 618(7963):126–133, 2023

    work page 2023

  3. [3]

    Generalizable spelling using a speech neuroprosthesis in an individual with severe limb and vocal paralysis.Nature communications, 13(1):6510, 2022

    Sean L Metzger, Jessie R Liu, David A Moses, Maximilian E Dougherty, Margaret P Seaton, Kaylo T Littlejohn, Josh Chartier, Gopala K Anumanchipalli, Adelyn Tu-Chan, Karunesh Ganguly, et al. Generalizable spelling using a speech neuroprosthesis in an individual with severe limb and vocal paralysis.Nature communications, 13(1):6510, 2022

    work page 2022

  4. [4]

    A streaming brain-to-voice neuroprosthesis to restore naturalistic communication.Nature neuroscience, 28(4):902–912, 2025

    Kaylo T Littlejohn, Cheol Jun Cho, Jessie R Liu, Alexander B Silva, Bohan Yu, Vanessa R Anderson, Cady M Kurtz-Miott, Samantha Brosler, Anshul P Kashyap, Irina P Hallinan, et al. A streaming brain-to-voice neuroprosthesis to restore naturalistic communication.Nature neuroscience, 28(4):902–912, 2025

    work page 2025

  5. [5]

    ECoG-based movement classification and limbs 3D translation prediction: a deep learning study

    Quentin Ferdinand, Rémi Souriau, Lucas Struber, Henri Lorach, Philippe Ciuciu, Marina Reyboz, and Tetiana Aksenova. ECoG-based movement classification and limbs 3D translation prediction: a deep learning study. In2025 International Joint Conference on Neural Networks (IJCNN), pages 1–10. IEEE, 2025

    work page 2025

  6. [6]

    Sandino, and Joseph Y

    Hsiang-Yun Sherry Chien, Hanlin Goh, Christopher M. Sandino, and Joseph Y . Cheng. Maeeg: Masked auto-encoder for eeg representation learning, 2022. URL https://arxiv.org/abs/ 2211.02625

    work page arXiv 2022

  7. [7]

    arXiv preprint arXiv:2305.10351 , year=

    Chaoqi Yang, M. Brandon Westover, and Jimeng Sun. BIOT: Cross-data biosignal learning in the wild, 2023. URLhttps://arxiv.org/abs/2305.10351

    work page arXiv 2023

  8. [8]

    arXiv preprint arXiv:2405.18765 , year=

    Wei-Bang Jiang, Li-Ming Zhao, and Bao-Liang Lu. Large brain model for learning generic representations with tremendous EEG data in BCI, 2024. URL https://arxiv.org/abs/ 2405.18765

    work page arXiv 2024

  9. [9]

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

    Jiquan Wang, Sha Zhao, Zhiling Luo, Yangxuan Zhou, Haiteng Jiang, Shijian Li, Tao Li, and Gang Pan. CBraMod: A criss-cross brain foundation model for EEG decoding, 2025. URL https://arxiv.org/abs/2412.07236

    work page arXiv 2025

  10. [10]

    Van Hulle

    Liuyin Yang, Qiang Sun, Ang Li, and Marc M. Van Hulle. Are EEG foundation models worth it? comparative evaluation with traditional decoders in diverse BCI tasks. InThe Fourteenth International Conference on Learning Representations, 2026. URL https://openreview. net/forum?id=5Xwm8e6vbh

    work page 2026

  11. [11]

    Induced electrocorticographic gamma activity during auditory perception.Clinical neurophysiology, 112(4):565–582, 2001

    Nathan E Crone, Dana Boatman, Barry Gordon, and Lei Hao. Induced electrocorticographic gamma activity during auditory perception.Clinical neurophysiology, 112(4):565–582, 2001

    work page 2001

  12. [12]

    Large EEG foundation model learns informative low-frequency representations from intracranial brain signals

    Qiang Sun, Liuyin Yang, Eva Calvo Merino, and Marc M Van Hulle. Large EEG foundation model learns informative low-frequency representations from intracranial brain signals. In1st ICLR Workshop on Time Series in the Age of Large Models, 2026

    work page 2026

  13. [13]

    Decoding flexion of individual fingers using electrocorticographic signals in humans.Journal of neural engineering, 6(6):066001, 2009

    JOJWGSJ Kubanek, Kai J Miller, Jeffrey G Ojemann, Jonathan R Wolpaw, and Gerwin Schalk. Decoding flexion of individual fingers using electrocorticographic signals in humans.Journal of neural engineering, 6(6):066001, 2009. 10

    work page 2009

  14. [14]

    Decoding of finger trajectory from ECoG using deep learning.Journal of neural engineering, 15(3):036009, 2018

    Ziqian Xie, Odelia Schwartz, and Abhishek Prasad. Decoding of finger trajectory from ECoG using deep learning.Journal of neural engineering, 15(3):036009, 2018

    work page 2018

  15. [15]

    Fast and accurate decoding of finger movements from ECoG through riemannian features and modern machine learning techniques.Journal of Neural Engineering, 19(1):016037, 2022

    Lin Yao, Bingzhao Zhu, and Mahsa Shoaran. Fast and accurate decoding of finger movements from ECoG through riemannian features and modern machine learning techniques.Journal of Neural Engineering, 19(1):016037, 2022

    work page 2022

  16. [16]

    Axel Faes, Flavio Camarrone, and Marc M Van Hulle. Single finger trajectory prediction from intracranial brain activity using block-term tensor regression with fast and automatic component extraction.IEEE Transactions on Neural Networks and Learning Systems, 35(7):8897–8908, 2022

    work page 2022

  17. [17]

    Towards real time efficient and robust ECoG decoding for mobile brain–computer interface.Journal of Neural Engineering, 22(4): 046012, 2025

    Zhanhui Lin, Xinyu Jiang, Chenyun Dai, and Fumin Jia. Towards real time efficient and robust ECoG decoding for mobile brain–computer interface.Journal of Neural Engineering, 22(4): 046012, 2025

    work page 2025

  18. [18]

    Simon, and Tom Francart

    Jonas Vanthornhout, Lien Decruy, Jan Wouters, Jonathan Z. Simon, and Tom Francart. Speech intelligibility predicted from neural entrainment of the speech envelope.Journal of the Associa- tion for Research in Otolaryngology, 19(2):181–191, 2018. doi: 10.1007/s10162-018-0654-z

    work page doi:10.1007/s10162-018-0654-z 2018

  19. [19]

    Decoding of the speech envelope from EEG using the VLAAI deep neural network.Scientific Reports, 13(1): 812, 2023

    Bernd Accou, Jonas Vanthornhout, Hugo Van hamme, and Tom Francart. Decoding of the speech envelope from EEG using the VLAAI deep neural network.Scientific Reports, 13(1): 812, 2023. doi: 10.1038/s41598-022-27332-2

    work page doi:10.1038/s41598-022-27332-2 2023

  20. [20]

    Decoding speech perception from non-invasive brain recordings.Nature Machine Intelligence, 5(10):1097–1107, 2023

    Alexandre Défossez, Charlotte Caucheteux, Jérémy Rapin, Ori Kabeli, and Jean-Rémi King. Decoding speech perception from non-invasive brain recordings.Nature Machine Intelligence, 5(10):1097–1107, 2023

    work page 2023

  21. [21]

    Ludovic Bellier, Anaïs Llorens, Déborah Marciano, Aysegul Gunduz, Gerwin Schalk, Peter Brunner, and Robert T. Knight. Music can be reconstructed from human auditory cortex activity using nonlinear decoding models.PLOS Biology, 21(8):e3002176, 2023. doi: 10.1371/journal. pbio.3002176

    work page doi:10.1371/journal 2023

  22. [22]

    Emergence of language in the developing brain, 2025

    Linnea Evanson, Christine Bulteau, Mathilde Chipaux, Georg Dorfmüller, Sarah Ferrand- Sorbets, Emmanuel Raffo, Sarah Rosenberg, Pierre Bourdillon, and Jean-Rémi King. Emergence of language in the developing brain, 2025. Preprint

    work page 2025

  23. [23]

    Transfer learning for brain–computer interfaces: A euclidean space data alignment approach.IEEE Transactions on Biomedical Engineering, 67(2):399–410, 2019

    He He and Dongrui Wu. Transfer learning for brain–computer interfaces: A euclidean space data alignment approach.IEEE Transactions on Biomedical Engineering, 67(2):399–410, 2019

    work page 2019

  24. [24]

    Generalized neural decoders for transfer learning across participants and recording modalities

    Steven M Peterson, Zoe Steine-Hanson, Nathan Davis, Rajesh PN Rao, and Bingni W Brunton. Generalized neural decoders for transfer learning across participants and recording modalities. Journal of Neural Engineering, 18(2):026014, 2021

    work page 2021

  25. [25]

    Methods for si- multaneous eeg-fmri: an introductory review.Journal of Neuroscience, 32(18):6053–6060, 2012

    René J Huster, Stefan Debener, Tom Eichele, and Christoph S Herrmann. Methods for si- multaneous eeg-fmri: an introductory review.Journal of Neuroscience, 32(18):6053–6060, 2012

    work page 2012

  26. [26]

    Richard N Henson, Daniel G Wakeman, Vladimir Litvak, and Karl J Friston. A parametric empirical bayesian framework for the eeg/meg inverse problem: generative models for multi- subject and multi-modal integration.Frontiers in human neuroscience, 5:76, 2011

    work page 2011

  27. [27]

    LoRA: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. LoRA: Low-rank adaptation of large language models. In International Conference on Learning Representations (ICLR), 2022

    work page 2022

  28. [28]

    Parameter-efficient transfer learning for nlp

    Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin De Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. Parameter-efficient transfer learning for nlp. InInternational conference on machine learning, pages 2790–2799. PMLR, 2019

    work page 2019

  29. [29]

    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. 11

    work page 2021

  30. [30]

    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

    work page 2020

  31. [31]

    On estimating regression.Theory of Probability & Its Applications, 9(1): 141–142, 1964

    Elizbar A Nadaraya. On estimating regression.Theory of Probability & Its Applications, 9(1): 141–142, 1964

    work page 1964

  32. [32]

    Smooth regression analysis.Sankhy ¯a: The Indian Journal of Statistics, Series A, pages 359–372, 1964

    Geoffrey S Watson. Smooth regression analysis.Sankhy ¯a: The Indian Journal of Statistics, Series A, pages 359–372, 1964

    work page 1964

  33. [33]

    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.arXiv preprint arXiv:2010.11929, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  34. [34]

    June 23, 2025.DOI:10.48550/arXiv.2506.19141

    Bruno Aristimunha, Dung Truong, Pierre Guetschel, Seyed Yahya Shirazi, Isabelle Guyon, Alexandre R. Franco, Michael P. Milham, Aviv Dotan, Scott Makeig, Alexandre Gramfort, Jean-Remi King, Marie-Constance Corsi, Pedro A. Valdés-Sosa, Amit Majumdar, Alan Evans, Terrence J Sejnowski, Oren Shriki, Sylvain Chevallier, and Arnaud Delorme. Eeg foundation challe...

    work page arXiv 2025

  35. [35]

    A library of human electrocorticographic data and analyses.Nature human behaviour, 3(11):1225–1235, 2019

    Kai J Miller. A library of human electrocorticographic data and analyses.Nature human behaviour, 3(11):1225–1235, 2019

    work page 2019

  36. [36]

    Spectro-temporal fusion of high-gamma and low-frequency ecog signals for intracranial finger movement decoding.TechRxiv, 2025

    Qiang Sun, Eva Calvo Merino, Bob Van Dyck, Yuan Yang, Jiayuan He, and Marc M Van Hulle. Spectro-temporal fusion of high-gamma and low-frequency ecog signals for intracranial finger movement decoding.TechRxiv, 2025

    work page 2025

  37. [37]

    Deepfingernet predicts finger trajectory from ecog measurements.IEEE Transactions on Instrumentation and Measurement, 74:1–12, 2025

    Wei Tao, Jianghao Hou, Yi Yang, Xun Chen, Tzyy-Ping Jung, and Feng Wan. Deepfingernet predicts finger trajectory from ecog measurements.IEEE Transactions on Instrumentation and Measurement, 74:1–12, 2025

    work page 2025

  38. [38]

    Mantisv2: Closing the zero-shot gap in time series classification with synthetic data and test-time strategies,

    Vasilii Feofanov, Songkang Wen, Jianfeng Zhang, Lujia Pan, and Ievgen Redko. Mantisv2: Closing the zero-shot gap in time series classification with synthetic data and test-time strategies,

    work page

  39. [39]

    URLhttps://arxiv.org/abs/2602.17868

    work page arXiv

  40. [40]

    The elements of statistical learning: data mining, inference, and prediction, 2009

    Trevor Hastie. The elements of statistical learning: data mining, inference, and prediction, 2009

    work page 2009

  41. [41]

    Demixed principal component analysis of neural population data.elife, 5:e10989, 2016

    Dmitry Kobak, Wieland Brendel, Christos Constantinidis, Claudia E Feierstein, Adam Kepecs, Zachary F Mainen, Xue-Lian Qi, Ranulfo Romo, Naoshige Uchida, and Christian K Machens. Demixed principal component analysis of neural population data.elife, 5:e10989, 2016

    work page 2016

  42. [42]

    Human motor cortical activity is selectively phase-entrained on underlying rhythms.PLoS Computational Biology, 8(9), 2012

    Kai J Miller, Dora Hermes, Christopher J Honey, Adam O Hebb, Nick F Ramsey, Robert T Knight, Jeffrey G Ojemann, and Eberhard E Fetz. Human motor cortical activity is selectively phase-entrained on underlying rhythms.PLoS Computational Biology, 8(9), 2012

    work page 2012

  43. [43]

    Human Motor Cortical Activity Is Selectively Phase- Entrained on Underlying Rhythms

    Peter L. Søndergaard and Piotr Majdak. The auditory modeling toolbox. In Jens Blauert, editor, The Technology of Binaural Listening, Modern Acoustics and Signal Processing, pages 33–56. Springer, Berlin, Heidelberg, 2013. 12 A Appendix A.1 LOO-FT Strategy Sweep and Failed Alternatives Loss and metric.CORTEG minimizes per-subject z-scored MSE L= 1 N NX i=1...

    work page arXiv 2013