arxiv: 2605.10817 · v1 · submitted 2026-05-11 · 💻 cs.AI

Recognition: 2 theorem links

· Lean Theorem

CLEF: EEG Foundation Model for Learning Clinical Semantics

Peng Cao , Ali Mirzazadeh , Jong Woo Lee , Aleksandar Videnovic , Dina Katabi

Authors on Pith no claims yet

Pith reviewed 2026-05-12 03:55 UTC · model grok-4.3

classification 💻 cs.AI

keywords EEGfoundation modelclinical semanticscontrastive learningspectrogram tokenslong-context modelingEHR alignmentneurologist reports

0 comments

The pith

CLEF aligns full-session EEG signals with neurologist reports and EHR via contrastive learning to learn transferable clinical semantics.

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

The paper presents CLEF as a foundation model that handles entire EEG sessions rather than short snippets, by turning recordings into 3D multitaper spectrogram tokens that a transformer can process at scale. It then uses contrastive objectives to pull these tokens close to embeddings from neurologist-written reports and structured electronic health record data. This design targets the gap where existing EEG models decode signals in isolation without clinical context. Tested across 234 tasks on disease labels, medication effects, and EEG abnormalities from over 260,000 sessions, the aligned model raises average AUROC from 0.65 to 0.74 and beats earlier foundation models on nearly all tasks. Even the version trained only on signal reconstruction already surpasses priors, while the clinical alignment adds measurable further gains and shows signs of working on held-out concepts and outside hospitals.

Core claim

CLEF represents EEG sessions as 3D multitaper spectrogram tokens for tractable long-context transformer modeling and aligns the resulting embeddings with neurologist reports and structured EHR data through contrastive objectives, producing representations that outperform prior EEG foundation models on 229 of 234 clinical tasks and transfer beyond the exact alignment targets used in training.

What carries the argument

3D multitaper spectrogram tokens for session-scale modeling combined with contrastive alignment to neurologist reports and EHR data.

If this is right

Representations support prediction of disease phenotypes from raw EEG sessions without task-specific labels.
The same embeddings improve detection of medication exposures and specific EEG findings across a large patient population.
Held-out concept experiments indicate the learned features apply to clinical targets never seen during alignment.
External-cohort results suggest the approach can generalize across different recording sites and patient groups.

Where Pith is reading between the lines

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

Similar alignment techniques could be applied to other long biosignal recordings such as overnight polysomnography when paired with clinical notes.
The session-scale tokenization may reduce the need for hand-crafted features in automated EEG review pipelines.
If the representations prove robust, they could serve as a backbone for multimodal models that jointly reason over EEG, imaging, and lab data.

Load-bearing premise

That matching EEG signals to reports and EHR through contrastive learning captures genuine transferable clinical meaning instead of dataset-specific statistical shortcuts.

What would settle it

Performance gains disappear when the model is tested on a new external hospital cohort that provides no report or EHR text at either training or test time, while a pure reconstruction version performs no better than earlier models.

Figures

Figures reproduced from arXiv: 2605.10817 by Aleksandar Videnovic, Ali Mirzazadeh, Dina Katabi, Jong Woo Lee, Peng Cao.

**Figure 1.** Figure 1: Model Performance. CLEF downstream performance across categories. Probing performance for top 5 downstream-task are averaged within each category. Detailed tasks are listed in Appendix E. are assessed not only by the morphology of individual discharges, but also by their recurrence and evolution. Encephalopathy, medication effects, and systemic illness often appear as broader changes in background organiza… view at source ↗

**Figure 2.** Figure 2: CLEF Pipeline. and denote trainable and frozen components, respectively. Stage I (left) pretrains the EEG encoder by masked reconstruction over a sequence of multi-channel spectro-temporal codes produced by a frozen tokenizer, with a parallel patch-embedding branch reinjecting details lost to quantization. Stage II (right) grounds the resulting patient-level embedding in clinical semantics by aligning it w… view at source ↗

**Figure 3.** Figure 3: Task probing performance across the benchmark. AUROC for CLEF (full and reconstruction-only variants) and five EEG foundation model baselines on (a) top 50 disease tasks, (b) top 50 medication tasks (c) all EEG feature tasks, sorted by CLEF performance. (d) AUROC across tasks, within each category and overall. clear from the EEG signal. This pattern is consistent with clinical practice: EEG is commonly use… view at source ↗

read the original abstract

Clinical EEG interpretation requires reasoning over full EEG sessions and integrating signal patterns with clinical context. Existing EEG foundation models are largely designed for short-window decoding and do not incorporate clinical context. We introduce CLEF, a clinically grounded long-context EEG foundation model. CLEF represents EEG sessions as 3D multitaper spectrogram tokens, enabling tractable Transformer modeling at session scale, and aligns embeddings with neurologist reports and structured EHR data through contrastive objectives. We evaluate CLEF on a new 234-task benchmark spanning disease phenotypes, medication exposures, and EEG findings, with more than 260k EEG sessions from over 108k patients. CLEF outperforms prior EEG foundation models on 229 of 234 tasks, improving mean AUROC from 0.65 to 0.74. Reconstruction-only pretraining surpasses prior EEG foundation models, while report and EHR alignment yields further gains. Held-out concept and external-cohort experiments suggest that these representations transfer beyond observed alignment targets. These results support session-scale, clinically grounded representation learning as a promising foundation-model paradigm for clinical EEG.

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.

Desk Editor's Note private letter to a colleague

CLEF shows that full-session 3D spectrogram tokens plus contrastive alignment to reports and EHR can lift EEG task performance across a large benchmark, but the case for genuine semantic transfer still rests on splits that share the same clinical sources.

read the letter

The main takeaway is that this model gets measurable gains on 229 of 234 EEG tasks by moving beyond short windows and tying embeddings to real clinical text and records. Reconstruction pretraining alone beats earlier foundation models, and the added contrastive step pushes mean AUROC from 0.65 to 0.74 on data from over 260k sessions and 108k patients. The 3D multitaper spectrogram tokenization is the concrete technical step that makes session-scale transformers feasible without excessive compute. That part is new relative to the short-window priors cited in the abstract, and the held-out concept plus external-cohort checks are a reasonable first pass at testing whether the alignment generalizes past the training targets. The scale of the new 234-task benchmark is also useful on its own; anyone running EEG experiments will appreciate having a standardized multi-task suite to compare against. The soft spot is the risk that the reported lifts partly reflect dataset-specific correlations rather than transferable clinical semantics. The benchmark tasks and the alignment data come from the same patient population and recording sources, so even with held-out splits there could be leakage through demographics, hardware, or correlated EHR fields. The external cohort is helpful, but the abstract does not spell out how different the protocols or populations actually are. Without seeing the exact split construction, statistical tests, and ablation tables, it is difficult to judge how robust the AUROC improvements are. This paper is aimed at researchers building clinical EEG models or multimodal foundation models who need a stronger baseline and a large evaluation set. A reader focused on practical neurophysiology AI will get value from the architecture and the benchmark even if the semantic claims require more work. It deserves peer review because the empirical scope is substantial and the core technical choices are clearly motivated. I would send it to referees and ask them to examine the data splits, any potential confounds between alignment and evaluation tasks, and the details of the external cohort.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces CLEF, a long-context EEG foundation model that tokenizes full sessions as 3D multitaper spectrograms for Transformer processing and uses contrastive alignment with neurologist reports and structured EHR data. It reports results on a new 234-task benchmark spanning phenotypes, medications, and EEG findings across >260k sessions from >108k patients, claiming outperformance versus prior EEG foundation models on 229 tasks (mean AUROC rising from 0.65 to 0.74), with reconstruction pretraining already surpassing baselines and alignment providing further gains, plus evidence of transfer from held-out concept and external-cohort experiments.

Significance. If the empirical claims and generalization results hold after detailed validation, the work would mark a meaningful step toward clinically contextualized, session-scale EEG representations. The scale of the benchmark and the explicit use of real clinical text/EHR for alignment are strengths that could influence future foundation-model design in clinical neurophysiology.

major comments (2)

[Abstract] Abstract: the headline claim that CLEF outperforms priors on 229 of 234 tasks and that alignment yields further gains rests on AUROC numbers whose robustness cannot be assessed without the data splits, statistical tests, confidence intervals, or ablation tables that separate reconstruction-only pretraining from report/EHR alignment.
[Abstract] Abstract: the assertion that held-out concept and external-cohort experiments demonstrate transfer 'beyond observed alignment targets' is load-bearing for the central claim of learning transferable clinical semantics, yet the abstract provides no concrete description of how these cohorts differ from the training distribution in patient demographics, recording hardware, or correlated EHR fields; without such details the experiments cannot yet distinguish semantic capture from dataset-specific correlations.

minor comments (1)

[Abstract] Abstract: the phrase '3D multitaper spectrogram tokens' is introduced without a short parenthetical on token dimensionality or how the representation remains tractable for full-session Transformers.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the two major comments point-by-point below. Both concerns relate to the level of detail in the abstract; we will make targeted revisions to the abstract to improve accessibility while preserving the manuscript's existing technical content.

read point-by-point responses

Referee: [Abstract] Abstract: the headline claim that CLEF outperforms priors on 229 of 234 tasks and that alignment yields further gains rests on AUROC numbers whose robustness cannot be assessed without the data splits, statistical tests, confidence intervals, or ablation tables that separate reconstruction-only pretraining from report/EHR alignment.

Authors: The full manuscript already contains these elements: data splits are specified in the Methods, statistical tests and confidence intervals appear in the Results and supplementary tables, and ablation tables isolating reconstruction pretraining from report/EHR alignment are presented in Section 4.2 and Figure 3. We agree the abstract would benefit from a brief reference to this supporting material and will revise it to state that 'results include statistical tests, confidence intervals, and ablations separating pretraining stages.' revision: partial
Referee: [Abstract] Abstract: the assertion that held-out concept and external-cohort experiments demonstrate transfer 'beyond observed alignment targets' is load-bearing for the central claim of learning transferable clinical semantics, yet the abstract provides no concrete description of how these cohorts differ from the training distribution in patient demographics, recording hardware, or correlated EHR fields; without such details the experiments cannot yet distinguish semantic capture from dataset-specific correlations.

Authors: The main text describes the held-out concepts (absent from alignment training) and external cohort (different institution, demographics, and hardware) in the Experiments section. We acknowledge the abstract is too terse on this point and will add a concise clause summarizing the key distributional differences to better support the transfer claim. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected in derivation or evaluation chain

full rationale

The paper's core claims rest on standard reconstruction pretraining of 3D spectrogram tokens followed by contrastive alignment to independent external data (neurologist reports and structured EHR). Evaluation uses a 234-task benchmark with explicit held-out concept splits and external-cohort testing to probe transfer beyond observed alignment targets. No equations, loss definitions, or results are shown to reduce by construction to fitted inputs or self-citations; the alignment targets and benchmark tasks are described as distinct from the EEG signals themselves. The derivation chain therefore remains self-contained against external benchmarks and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The approach rests on standard assumptions of contrastive representation learning and spectrogram-based signal processing; no new physical entities or ad-hoc constants are introduced.

axioms (2)

domain assumption Multitaper spectrograms in 3D preserve the clinically relevant information in long EEG sessions.
Core representational choice stated in the abstract.
domain assumption Contrastive objectives between EEG embeddings and clinical text embeddings produce semantically meaningful alignments.
Central to the report and EHR alignment step.

pith-pipeline@v0.9.0 · 5493 in / 1422 out tokens · 134471 ms · 2026-05-12T03:55:52.459767+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

CLEF represents EEG sessions as 3D multitaper spectrogram tokens... aligns embeddings with neurologist reports and structured EHR data through contrastive objectives... Stage I: Session-Scale Masked Reconstruction... Stage II: Cross-Modal Alignment with Report and EHR

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

54 extracted references · 54 canonical work pages · 3 internal anchors

[1]

Harvard electroencephalography database: A comprehensive clinical electroencephalographic resource from four boston hospitals.Epilepsia, 66(9):3411–3425, 2025

C Sun, J Jing, N Turley, C Alcott, WY Kang, AJ Cole, DM Goldenholz, A Lam, E Amorim, C Chu, et al. Harvard electroencephalography database: A comprehensive clinical electroencephalographic resource from four boston hospitals.Epilepsia, 66(9):3411–3425, 2025

work page 2025
[2]

Bendr: Using transformers and a contrastive self-supervised learning task to learn from massive amounts of eeg data.Frontiers in Human Neuroscience, 15:653659, 2021

Demetres Kostas, Stephane Aroca-Ouellette, and Frank Rudzicz. Bendr: Using transformers and a contrastive self-supervised learning task to learn from massive amounts of eeg data.Frontiers in Human Neuroscience, 15:653659, 2021

work page 2021
[3]

Maeeg: Masked auto-encoder for eeg representation learning

Hsiang-Yun Sherry Chien, Hanlin Goh, Christopher Michael Sandino, and Joseph Yitan Cheng. Maeeg: Masked auto-encoder for eeg representation learning. InNeurIPS 2022 Workshop on Learning from Time Series for Health, 2022

work page 2022
[4]

Neuro-gpt: Towards a foundation model for eeg

Wenhui Cui, Woojae Jeong, Philipp Thölke, Takfarinas Medani, Karim Jerbi, Anand A Joshi, and Richard M Leahy. Neuro-gpt: Towards a foundation model for eeg. In2024 IEEE International Symposium on Biomedical Imaging (ISBI), pages 1–5. IEEE, 2024

work page 2024
[5]

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. InThe Twelfth International Conference on Learning Representations, 2024

work page 2024
[6]

Eegformer: Towards transferable and interpretable large-scale eeg foundation model

Yuqi Chen, Kan Ren, Kaitao Song, Yansen Wang, Yifan Wang, Dongsheng Li, and Lili Qiu. Eegformer: Towards transferable and interpretable large-scale eeg foundation model. InAAAI 2024 Spring Symposium on Clinical Foundation Models

work page 2024
[7]

CBramod: A criss-cross brain foundation model for EEG decoding

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. InThe Thirteenth International Conference on Learning Representations, 2025

work page 2025
[8]

EEGPT: Pretrained trans- former for universal and reliable representation of EEG signals

Guangyu Wang, Wenchao Liu, Yuhong He, Cong Xu, Lin Ma, and Haifeng Li. EEGPT: Pretrained trans- former for universal and reliable representation of EEG signals. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

work page 2024
[9]

CSBrain: A cross-scale spatiotemporal brain foundation model for EEG decoding

Yuchen Zhou, Jiamin Wu, Zichen Ren, Zhouheng Yao, Weiheng Lu, Kunyu Peng, Qihao Zheng, Chunfeng Song, Wanli Ouyang, and Chao Gou. CSBrain: A cross-scale spatiotemporal brain foundation model for EEG decoding. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

work page 2025
[10]

THD-BAR: Topology hierarchical derived brain autoregressive modeling for EEG generic representations

Wenchao Yang, Weidong Yan, Wenkang Liu, Yulan Ma, and Yang Li. THD-BAR: Topology hierarchical derived brain autoregressive modeling for EEG generic representations. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

work page 2025
[11]

REVE: A foundation model for EEG - adapting to any setup with large-scale pretraining on 25,000 subjects

Yassine El Ouahidi, Jonathan Lys, Philipp Thölke, Nicolas Farrugia, Bastien Pasdeloup, Vincent Gripon, Karim Jerbi, and Giulia Lioi. REVE: A foundation model for EEG - adapting to any setup with large-scale pretraining on 25,000 subjects. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

work page 2025
[12]

Saurabh R Sinha, Lucy R Sullivan, Dragos Sabau, Daniel San Juan Orta, Keith E Dombrowski, Jonathan J Halford, Abeer J Hani, Frank W Drislane, and Mark M Stecker. American clinical neurophysiology society guideline 1: minimum technical requirements for performing clinical electroencephalography.The Neurodiagnostic Journal, 56(4):235–244, 2016

work page 2016
[13]

American clinical neurophysiology society guideline 7: guidelines for eeg reporting.The Neurodiagnostic Journal, 56(4):285–293, 2016

William O Tatum IV , Olga Selioutski, Juan G Ochoa, Heidi Munger Clary, Janna Cheek, Frank W Drislane, and Tammy N Tsuchida. American clinical neurophysiology society guideline 7: guidelines for eeg reporting.The Neurodiagnostic Journal, 56(4):285–293, 2016

work page 2016
[14]

MIT press, 2014

Mike X Cohen.Analyzing neural time series data: theory and practice. MIT press, 2014

work page 2014
[15]

Oxford university press, 2006

György Buzsáki.Rhythms of the Brain. Oxford university press, 2006

work page 2006
[16]

Oxford university press, 2007

Partha Mitra.Observed brain dynamics. Oxford university press, 2007

work page 2007
[17]

Spectrum estimation and harmonic analysis.Proceedings of the IEEE, 70(9):1055–1096, 1982

David J Thomson. Spectrum estimation and harmonic analysis.Proceedings of the IEEE, 70(9):1055–1096, 1982

work page 1982
[18]

Taming transformers for high-resolution image synthesis

Patrick Esser, Robin Rombach, and Bjorn Ommer. Taming transformers for high-resolution image synthesis. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 12873–12883, 2021. 10

work page 2021
[19]

Learning transferable visual models from natural language supervision

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. InInternational conference on machine learning, pages 8748–8763. PmLR, 2021

work page 2021
[20]

Contrastive learning of medical visual representations from paired images and text

Yuhao Zhang, Hang Jiang, Yasuhide Miura, Christopher D Manning, and Curtis P Langlotz. Contrastive learning of medical visual representations from paired images and text. InMachine learning for healthcare conference, pages 2–25. PMLR, 2022

work page 2022
[21]

Expert- level detection of pathologies from unannotated chest x-ray images via self-supervised learning.Nature biomedical engineering, 6(12):1399–1406, 2022

Ekin Tiu, Ellie Talius, Pujan Patel, Curtis P Langlotz, Andrew Y Ng, and Pranav Rajpurkar. Expert- level detection of pathologies from unannotated chest x-ray images via self-supervised learning.Nature biomedical engineering, 6(12):1399–1406, 2022

work page 2022
[22]

BiomedCLIP: a multimodal biomedical foundation model pretrained from fifteen million scientific image-text pairs

Sheng Zhang, Yanbo Xu, Naoto Usuyama, Hanwen Xu, Jaspreet Bagga, Robert Tinn, Sam Preston, Rajesh Rao, Mu Wei, Naveen Valluri, et al. Biomedclip: a multimodal biomedical foundation model pretrained from fifteen million scientific image-text pairs.arXiv preprint arXiv:2303.00915, 2023

work page internal anchor Pith review arXiv 2023
[23]

Sedigheh Eslami, Christoph Meinel, and Gerard De Melo. Pubmedclip: How much does clip benefit visual question answering in the medical domain? InFindings of the Association for Computational Linguistics: EACL 2023, pages 1181–1193, 2023

work page 2023
[24]

Open vocabulary electroencephalography-to-text decoding and zero-shot sentiment classification

Zhenhailong Wang and Heng Ji. Open vocabulary electroencephalography-to-text decoding and zero-shot sentiment classification. InProceedings of the AAAI Conference on Artificial Intelligence, volume 36, pages 5350–5358, 2022

work page 2022
[25]

Dewave: Discrete encoding of eeg waves for eeg to text translation.Advances in Neural Information Processing Systems, 36:9907–9918, 2023

Yiqun Duan, Jinzhao Zhou, Zhen Wang, Yu-Kai Wang, and Chin-teng Lin. Dewave: Discrete encoding of eeg waves for eeg to text translation.Advances in Neural Information Processing Systems, 36:9907–9918, 2023

work page 2023
[26]

Brainclip: Brain representation via clip for generic natural visual stimulus decoding.IEEE Transactions on Medical Imaging, 44(10):3962–3972, 2025

Yongqiang Ma, Yulong Liu, Liangjun Chen, Guibo Zhu, Badong Chen, and Nanning Zheng. Brainclip: Brain representation via clip for generic natural visual stimulus decoding.IEEE Transactions on Medical Imaging, 44(10):3962–3972, 2025

work page 2025
[27]

NeuroLM: A universal multi-task foundation model for bridging the gap between language and EEG signals

Weibang Jiang, Yansen Wang, Bao liang Lu, and Dongsheng Li. NeuroLM: A universal multi-task foundation model for bridging the gap between language and EEG signals. InThe Thirteenth International Conference on Learning Representations, 2025

work page 2025
[28]

Exploring the limits of transfer learning with a unified text-to-text transformer

Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research, 21(140):1–67, 2020

work page 2020
[29]

Scaling language-image pre-training via masking

Yanghao Li, Haoqi Fan, Ronghang Hu, Christoph Feichtenhofer, and Kaiming He. Scaling language-image pre-training via masking. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 23390–23400, 2023

work page 2023
[30]

Systematic comparison of phenome- wide association study of electronic medical record data and genome-wide association study data.Nature biotechnology, 31(12):1102–1111, 2013

Joshua C Denny, Lisa Bastarache, Marylyn D Ritchie, Robert J Carroll, Raquel Zink, Jonathan D Mosley, Julie R Field, Jill M Pulley, Andrea H Ramirez, Erica Bowton, et al. Systematic comparison of phenome- wide association study of electronic medical record data and genome-wide association study data.Nature biotechnology, 31(12):1102–1111, 2013

work page 2013
[31]

Atc/ddd index

WHO Collaborating Centre for Drug Statistics Methodology. Atc/ddd index. URL https://atcddd. fhi.no/atc_ddd_index/

work page
[32]

The temple university hospital eeg data corpus.Frontiers in neuroscience, 10:196, 2016

Iyad Obeid and Joseph Picone. The temple university hospital eeg data corpus.Frontiers in neuroscience, 10:196, 2016

work page 2016
[33]

Screening for cognitive impairment during sleep studies: The George B

PhysioNet Challenge. Screening for cognitive impairment during sleep studies: The George B. Moody PhysioNet Challenge 2026. https://moody-challenge.physionet.org/2026/, 2026. Data sourced from the Human Sleep Project (HSP)

work page 2026
[34]

BIOT: Biosignal transformer for cross-data learning in the wild

Chaoqi Yang, M Brandon Westover, and Jimeng Sun. BIOT: Biosignal transformer for cross-data learning in the wild. InThirty-seventh Conference on Neural Information Processing Systems, 2023

work page 2023
[35]

Neurolm: A universal multi- task foundation model for bridging the gap between lan- guage and eeg signals.arXiv preprint arXiv:2409.00101,

Wei-Bang Jiang, Yansen Wang, Bao-Liang Lu, and Dongsheng Li. Neurolm: A universal multi-task foundation model for bridging the gap between language and eeg signals, 2025. URL https://arxiv. org/abs/2409.00101. 11

work page arXiv 2025
[36]

Ruijter, Marleen C

Barry J. Ruijter, Marleen C. Tjepkema-Cloostermans, Selma C. Tromp, Walter M. van den Bergh, Norbert A. Foudraine, Francois H. M. Kornips, Gea Drost, Erik Scholten, Frank H. Bosch, Albertus Beishuizen, Michel J. A. M. van Putten, and Jeannette Hofmeijer. Early electroencephalography for outcome prediction of postanoxic coma: A prospective cohort study.Ann...

work page doi:10.1002/ana.25518 2019
[37]

Purdon, Eric T

Patrick L. Purdon, Eric T. Pierce, Eran A. Mukamel, Michael J. Prerau, John L. Walsh, Kin Foon K. Wong, Andres F. Salazar-Gomez, Priscilla G. Harrell, Aaron L. Sampson, Aylin Cimenser, ShiNung Ching, Nancy J. Kopell, Casie Tavares-Stoeckel, Kathleen Habeeb, Rebecca Merhar, and Emery N. Brown. Electroencephalogram signatures of loss and recovery of conscio...

work page doi:10.1073/pnas.1221180110 2013
[38]

Aton, Julie Seibt, Marie C

Sara J. Aton, Julie Seibt, Marie C. Dumoulin, Timothy Coleman, Masahiro Shiraishi, and Marcos G. Frank. The sedating antidepressant trazodone impairs sleep-dependent cortical plasticity.PLOS ONE, 4(7): e6078, 2009. doi: 10.1371/journal.pone.0006078. URL https://doi.org/10.1371/journal.pone. 0006078

work page doi:10.1371/journal.pone.0006078 2009
[39]

Boyd, Janice Brunstrom-Hernandez, Giovanni Cioni, Diane Damiano, Johanna Darrah, Ann-Christin Eliasson, Linda S

Iona Novak, Cathy Morgan, Lars Adde, James Blackman, Roslyn N. Boyd, Janice Brunstrom-Hernandez, Giovanni Cioni, Diane Damiano, Johanna Darrah, Ann-Christin Eliasson, Linda S. de Vries, Christa Einspieler, Michael Fahey, Darcy Fehlings, Donna M. Ferriero, Linda Fetters, Simona Fiori, Hans Forssberg, Andrew M. Gordon, Susan Greaves, Andrea Guzzetta, Mijna ...

work page doi:10.1001/jamapediatrics 2017
[40]

Qwen2 Technical Report

An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jin Xu, Jingren Zhou, Jinze Bai, Jinzheng He, Junyang Lin, Kai Dang, Keming Lu, Keqin Chen, Kexin Yang, Mei Li, Mingfeng ...

work page internal anchor Pith review Pith/arXiv arXiv 2024
[41]

Qwen2.5: A party of foundation models, September 2024

Qwen Team. Qwen2.5: A party of foundation models, September 2024. URL https://qwenlm.github. io/blog/qwen2.5/

work page 2024
[42]

Hermes-2-pro-llama-3-8b, 2024

Teknium, interstellarninja, theemozilla, karan4d, and huemin_art. Hermes-2-pro-llama-3-8b, 2024. URL https://huggingface.co/NousResearch/Hermes-2-Pro-Llama-3-8B. NousResearch

work page 2024
[43]

Aasm scoring manual updates for 2017 (version 2.4), 2017

Richard B Berry, Rita Brooks, Charlene Gamaldo, Susan M Harding, Robin M Lloyd, Stuart F Quan, Matthew T Troester, and Bradley V Vaughn. Aasm scoring manual updates for 2017 (version 2.4), 2017

work page 2017
[44]

Meg and eeg data analysis with mne-python.Frontiers in Neuroinformatics, 7:267, 2013

Alexandre Gramfort, Martin Luessi, Eric Larson, Denis A Engemann, Daniel Strohmeier, Christian Brodbeck, Roman Goj, Mainak Jas, Teon Brooks, Lauri Parkkonen, et al. Meg and eeg data analysis with mne-python.Frontiers in Neuroinformatics, 7:267, 2013

work page 2013
[45]

Braindecode: toolbox for decoding raw electrophysiological brain data with deep learning models

Bruno Aristimunha, Pierre Guetschel, Martin Wimpff, Lukas Gemein, Cedric Rommel, Hubert Banville, Maciej Sliwowski, Daniel Wilson, Simon Brandt, Théo Gnassounou, Joseph Paillard, Bruna Junqueira Lopes, Sara Sedlar, Thomas Moreau, Sylvain Chevallier, Alexandre Gramfort, and Robin Tibor Schirrmeis- ter. Braindecode: toolbox for decoding raw electrophysiolog...

work page
[46]

Deep learning with convolutional neural networks for eeg decoding and visualization.Human Brain Mapping, aug

Robin Tibor Schirrmeister, Jost Tobias Springenberg, Lukas Dominique Josef Fiederer, Martin Glasstetter, Katharina Eggensperger, Michael Tangermann, Frank Hutter, Wolfram Burgard, and Tonio Ball. Deep learning with convolutional neural networks for eeg decoding and visualization.Human Brain Mapping, aug

work page
[47]

doi: 10.1002/hbm.23730

ISSN 1097-0193. doi: 10.1002/hbm.23730. URLhttp://dx.doi.org/10.1002/hbm.23730

work page doi:10.1002/hbm.23730
[48]

The Llama 3 Herd of Models

Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al- Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. The llama 3 herd of models. arXiv preprint arXiv:2407.21783, 2024. 12 A Discussion A.1 Limitations Lack of Real-time Support.CLEF summarizes an entire EEG session as a single patient-level v...

work page internal anchor Pith review Pith/arXiv arXiv 2024
[49]

Begin with: seizures, epileptiform discharges, focal abnormalities, significant slowing

work page
[50]

Continue with: background rhythm description, frequency, amplitude, organization, symmetry

work page
[51]

Then add: sleep stages, reactivity, activation results

work page
[52]

Then add: clinical impression, diagnostic significance

work page
[53]

Then add: study type, duration, quality indicators

work page
[54]

PRN”, or “take as needed

End with: intracranial EEG data (scalp EEG takes precedence) DELETION DIRECTIVE - Eliminate: Timestamps | Electrode names/positions | Format markup | Duplicate phrases | Prior study references PRESERVATION DIRECTIVE - Keep intact: Medical terminology (exact) | Negations (no/without/absence) | Quantifiers (mild/moderate/severe/frequent/rare) | Medications ...

work page arXiv 2048