[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-125319-en":3,"doc-seo-125319-105":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":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},125319,962075114765,"Quinn","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","Efficient Personalized Adaptation for Physiological Signal Foundation Model","Time series analysis underpins healthcare tasks such as monitoring and diagnosis, yet existing deep models struggle with limited local compute and strict medical data privacy. Although Time Series Foundation Models (TSFM) benefit from large-scale pre-training, generic TSFM can underperform on private, imbalanced physiological data. This work proposes PhysioPFM to efficiently personalize TSFM for medical practice, using low-rank pre-training on public sets, generator training with trained LoRA weights, and weight generation driven by local data. Results show improved performance with generated models, better transferability, and reduced need for additional sensitive training.","Efficient Personalized Adaptation for Physiological Signal Foundation Model  \nChenrui Wu 1 2 Haishuai Wang∗ 1 Xiang Zhang 3 Chengqi Zhang 4 Jiajun Bu 1  \nAbstract  \nTime series analysis is crucial across various fields like energy, environment, transportation, finance and health. Deep learning has significantly advanced this field, particularly, the Time Series Foundation Model (TSFM) excels in multiple domains due to extensive pre-training. In this work, we focus on TSFM’s challenges in medical practice: limited computing resources and medical data privacy. TSFM variants include fine-tuned models and those pre-trained for rapid deployment on diverse data. There may not be enough computing resources to train physiological signals locally in hospitals, and generalized TSFM is still inferior to task-specific methods on private, imbalanced local data. To address this, we propose PhysioPFM, a framework for efficiently personalizing TSFM. Our approach involves low-rank pre-training on public datasets, generator training by trained LoRA weights, and efficient weight generation via local data. Experimental results demonstrate that integrating generated models with TSFM enhances performance, and transferability, and reduces the need for additional sensitive data training.  \n1. Introduction  \nThe recent trend of integrating deep learning algorithmson advanced wearable sensors and fixed medical equipment has catalyzed massive amounts of valuable medical data (Spathis et al., 2020 ; Che et al., 2017) . These recorded medical time series data are continuous observations related to human health, usually including electroencephalogram (EEG), heart rate, cardiotocography (CTG), electrooculogram (EOG), galvanic skin response (GSR), electrocardiogram (ECG), electromyogram (EMG) and others (Thapa  \n1Zhejiang University. 2 Simon Fraser University. 3The University of North Carolina at Charlotte. 4Hong Kong Polytechnic University. Correspondence to: Haishuai Wang \u003C[haishuai.wang@zju.edu.cn](haishuai.wang@zju.edu.cn) >.  \nProceedings of the 42 nd International Conference on Machine Learning, Vancouver, Canada. PMLR 267, 2025 . Copyright 2025 by the author(s) .  \n(a) TSFM for Local Adaption (b) TSFM for Generic Pretraining  \nFigure 1 . Illustration of existing two category time series foundation model. (a) represents the methods applying pre-trained large language model to train target data, and conducting testson target data. (b) demonstrate the methods that pre-train on a large and multi-domain dataset, then test on the target data for the zero-shot prediction.  \net al., 2024 ; Wang et al., 2024b ; Zhang et al., 2024 ; Al-Saeghet al., 2021) . These physiological signals are usually quantitatively measured by medical devices and then analyzed by doctors or specialists to evaluate the patient’s current state and make data-driven decisions, which shows great significance for health monitoring, disease diagnosis and treatment (Liu et al., 2023) . Specifically, we focus on various medical time series classification (TSC) tasks based on the physiological signals, including emotion recognition (Pan et al., 2023 ; Li et al., 2024a), sleep stage detection (Suprataket al., 2017 ; Dong et al., 2017), neurological disorder classification (Yang et al., 2022b ; Zhang et al., 2022) etc.  \nHowever, while the community has benefited greatly from the large amount of new data collected by professional medical devices or ubiquitous wearable devices, analyzing physiological signals with existing deep-learning methods still faces inherent challenges in practice. First, The amount of data available for each signal is unbalanced. Most existing studies focus on EEG and ECG data. In contrast, other physiological signals have minor data available, making it challenging to establish unified and generic models for all signals. Furthermore, the sampling frequency and duration of different signals may also vary, further causing the divergence of data features and labels. Theref","cbCaijVXr3EEhLqA","https://ap.wps.com/l/cbCaijVXr3EEhLqA","pdf",1181724,1,19,"English","en",105,"# Introduction\n## Medical time series classification tasks\n## Challenges in physiological signal learning\n## Existing TSFM categories\n## Motivation for efficient personalization","[{\"question\":\"Why is personalizing a time series foundation model important in physiological signal tasks?\",\"answer\":\"Physiological data are private and often imbalanced across patients and devices, and localized clinical scenarios may not support heavy fine-tuning. Generic TSFM can also underperform on specific tasks with scarce public healthcare data.\"},{\"question\":\"What challenges does the proposed PhysioPFM aim to solve?\",\"answer\":\"It targets limited computing resources for local hospital deployment and medical data privacy that prevents uploading sensitive signals for pre-training. It also addresses performance gaps caused by private, imbalanced local datasets.\"},{\"question\":\"How does PhysioPFM perform efficient personalization of TSFM?\",\"answer\":\"PhysioPFM uses low-rank pre-training on public datasets, trains a generator guided by trained LoRA weights, and generates efficient weights using only local data. The generated models are then integrated with TSFM for improved results.\"}]","Efficient Personalized Adaptation for Physiological Signal Foundation Model | PDF",1785898155,48,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"efficient-personalized-adaptation-for-physiological-signal-foundation-model","",{"@graph":36,"@context":85},[37,54,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"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":53},"https://docshare.wps.com/document/efficient-personalized-adaptation-for-physiological-signal-foundation-model/125319/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":62,"encodingFormat":61,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-05",true,{"@type":65,"interactionType":66,"userInteractionCount":4},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why is personalizing a time series foundation model important in physiological signal tasks?","Question",{"text":75,"@type":76},"Physiological data are private and often imbalanced across patients and devices, and localized clinical scenarios may not support heavy fine-tuning. Generic TSFM can also underperform on specific tasks with scarce public healthcare data.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What challenges does the proposed PhysioPFM aim to solve?",{"text":80,"@type":76},"It targets limited computing resources for local hospital deployment and medical data privacy that prevents uploading sensitive signals for pre-training. It also addresses performance gaps caused by private, imbalanced local datasets.",{"name":82,"@type":73,"acceptedAnswer":83},"How does PhysioPFM perform efficient personalization of TSFM?",{"text":84,"@type":76},"PhysioPFM uses low-rank pre-training on public datasets, trains a generator guided by trained LoRA weights, and generates efficient weights using only local data. The generated models are then integrated with TSFM for improved results.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"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":53,"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":21,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":106,"slug":137},"General","general"]