[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86270-en":3,"doc-seo-86270-105":30,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},86270,1099514068365,"Aurelia","https://ap-avatar.wpscdn.com/avatar/10000253d8d9f28188e?_k=1776742907772140068",8,"Research & Report","DiffEEG A Self-Supervised Denoising Diffusion Model for Learning EEG Generic Representations","Deep learning for EEG-based seizure detection faces severe annotation scarcity and extreme class imbalance, with ictal events under 10% of clinical recordings. DiffEEG introduces a 9.6M-parameter self-supervised foundation model combining denoising diffusion pre-training and reinforcement learning fine-tuning. Pre-trained on 1.3M unlabeled TUHSZ segments using a 1D U-Net with multi-head self-attention, it adapts with a policy-gradient decision layer that directly maximizes F1-score for rare-event sensitivity. Patient-wise Leave-One-Fold-Out evaluation yields strong multiclass and binary performance with clinically viable recall, while segment-level results confirm architectural capacity and diffusion-based metric-aware training.","arXiv :2607 . 1 1578v 1 [ cs .LG] 13 Jul 2026  \nDiffEEG: A Self-Supervised Denoising Diffusion Model for Learning EEG Generic Representations  \nAbdulkader Helwan 1, Lina Abou-Abbas 1 ,∗, Hussein El Amouri 1, Belkacem Chikhaoui2 and Khadidja Henni2  \n1 Department of Electrical and Computer Engineering, Lebanese American University, Byblos, Lebanon  \n2 Institute of Applied Artificial Intelligence, T´ELUQ University, Montreal, Canada  \n∗ Author to whom any correspondence should be addressed.  \nE-mail: [lina.abouabbas@lau.edu.lb](lina.abouabbas@lau.edu.lb)  \nKeywords: EEG, Deep learning, denoising diffusion, foundation model, reinforcement learning.  \nAbstract  \nDeep learning for EEG-based seizure detection faces critical challenges: severe annotation scarcity and extreme class imbalance, where ictal events comprise less than 10% of clinical recordings. We present DiffEEG, a 9.6M-parameter self-supervised foundation model that addresses both limitations through denoising diffusion pre-training and reinforcement learning (RL)-based fine-tuning. Pre-trained on 1.3M unlabeled segments from the Temple University Hospital Seizure Corpus (TUHSZ), DiffEEG learns generic neural representations via a 1D U-Net with multi-head self-attention. For downstream adaptation, a reinforced decision layer employs policy gradient optimization to directly maximize F1-score, prioritizing sensitivity to rare seizure events over overall accuracy. Under strict patient-wise evaluation (279 patients, Leave-One-Fold-Out), DiffEEG achieves 61% accuracy and 59% F1 for 4-class seizure subtyping, and 81% accuracy with 85% weighted F1 for binary detection, maintaining clinically viable seizure recall (59%) despite extreme imbalance (6.7% prevalence) . Segment-level evaluation establishes an upper bound of 97 .6% accuracy, confirming strong architectural capacity. DiffEEG demonstrates that diffusion-based pre-training combined with metric-aware reinforcement learning enables clinically deployable seizure monitoring with minimal labeled data requirements.  \n1 Introduction  \nEpilepsy is one of the most prevalent neurological disorders globally, characterized by recurrent, unprovoked electrical disturbances in the brain [1], affecting more than 50 million individuals of all ages worldwide. Electroencephalography (EEG) remains the primary non-invasive diagnostic tool for identifying seizures due to its high temporal resolution; however, the manual interpretation of EEG recordings is a labor-intensive and expensive process that necessitates highly specialized expertise [1–3], and visual analysis frequently suffers from significant inter-rater variability that compromises diagnostic consistency in acute clinical settings [4] . To overcome these constraints, researchers have increasingly employed machine learning and deep learning to identify latentspatio-temporal patterns in multi-channel EEG signals [2,3,5–10] . The field transitioned from handcrafted feature engineering toward automatic feature extraction through convolutional andrecurrent architectures [2,3,6,9,11], and more recently through transformer-based models capable of capturing long-range temporal dependencies via self-attention [12–15] . Multimodal strategies, such as EEG-ECG synchronization toward EEG-free detection [16], were also explored in parallel, though without resolving the fundamental limitation that generalization across patients, recording setups, and institutional protocols remained fragile, and the volume of labeled EEG required to train these models reliably remained a persistent practical barrier. These limitations motivated a shift toward EEG foundation models (FMs), large architectures pretrained in a self-supervised or unsupervised fashion on thousands of hours of unlabeled recordings from corpora such as the Temple University Hospital (TUH) EEG Corpus [17], and subsequently fine-tuned on specific downstream tasks [5,12,18–25] . By learning universal brain activity representations from mas","cbCaiihEYIoo4wkV","https://ap.wps.com/l/cbCaiihEYIoo4wkV","pdf",916069,4,1,19,"English","en",105,"# Introduction\n## EEG for seizure detection challenges\n## Foundation models for EEG representation learning\n## Limitations of existing benchmarks\n## Motivation and proposed approach","[{\"question\":\"What problem does DiffEEG aim to solve in EEG seizure detection?\",\"answer\":\"DiffEEG targets annotation scarcity and extreme class imbalance, where seizure (ictal) events make up less than 10% of clinical EEG recordings.\"},{\"question\":\"How is DiffEEG trained to learn generic EEG representations?\",\"answer\":\"It uses denoising diffusion self-supervised pre-training on 1.3M unlabeled TUHSZ segments with a 1D U-Net and multi-head self-attention.\"},{\"question\":\"How does DiffEEG adapt to downstream tasks during fine-tuning?\",\"answer\":\"A reinforced decision layer applies policy-gradient optimization to directly maximize F1-score, emphasizing sensitivity to rare seizure events rather than overall accuracy.\"}]",1784209942,48,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"diffeeg-a-self-supervised-denoising-diffusion-model-for-learning-eeg-generic-representations","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":20},"https://docshare.wps.com/document/diffeeg-a-self-supervised-denoising-diffusion-model-for-learning-eeg-generic-representations/86270/",{"url":52,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-25","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does DiffEEG aim to solve in EEG seizure detection?","Question",{"text":75,"@type":76},"DiffEEG targets annotation scarcity and extreme class imbalance, where seizure (ictal) events make up less than 10% of clinical EEG recordings.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How is DiffEEG trained to learn generic EEG representations?",{"text":80,"@type":76},"It uses denoising diffusion self-supervised pre-training on 1.3M unlabeled TUHSZ segments with a 1D U-Net and multi-head self-attention.",{"name":82,"@type":73,"acceptedAnswer":83},"How does DiffEEG adapt to downstream tasks during fine-tuning?",{"text":84,"@type":76},"A reinforced decision layer applies policy-gradient optimization to directly maximize F1-score, emphasizing sensitivity to rare seizure events rather than overall accuracy.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":22,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":106,"slug":137},"General","general"]