[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-121505-en":3,"doc-seo-121505-105":30,"detail-sidebar-cat-0-en-105":95},{"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":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},121505,1374391974585,"Genevieve","https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c",8,"Research & Report","Machine Learning in Resource-Constrained Distributed and Dynamic Environments - Dissertation","Modern machine learning models deliver strong performance but face major deployment barriers when systems are distributed, dynamic, and strictly resource-limited. This dissertation proposes new algorithms and theoretical analyses to close efficiency and adaptability gaps. It introduces AdMiT for adaptive parameter-efficient tuning on unlabeled edge data using kernel mean embedding module matching. It also addresses asynchronous federated learning under non-IID data by mitigating heterogeneity amplification with AC-AFL. Finally, it studies in-context learning with prompt-tuning in large language models, offering covariance-mean decoupling results.","UC Riverside  \nUC Riverside Electronic Theses and Dissertations  \nTitle  \nMachine Learning in Resource-Constrained Distributed and Dynamic Environments  \nPermalink  \n[https://escholarship.org/uc/item/0hp6q4wh](https://escholarship.org/uc/item/0hp6q4wh)  \nISBN  \n9798263308544  \nAuthor  \nChang, Xiangyu  \nPublication Date  \n2025-08-20  \nPeer reviewed|Thesis/dissertation  \n[eScholarship.org](eScholarship.org) Powered by the California Digital Library  \nUniversity of California  \nUNIVERSITY OF CALIFORNIA  \nRIVERSIDE  \nMachine Learning in Resource-Constrained Distributed and Dynamic Environments  \nA Dissertation submitted in partial satisfaction of the requirements for the degree of  \nDoctor of Philosophy  \nin  \nElectrical Engineering  \nby  \nXiangyu Chang  \nSeptember 2025  \nDissertation Committee:  \nProf. Amit K. Roy-Chowdhury, Chairperson  \nProf. Samet Oymak  \nProf. Srikanth V. Krishnamurthy  \nProf. Yinglun Zhu  \nDr. Ananthram Swami  \nCopyright by Xiangyu Chang 2025  \nThe Dissertation of Xiangyu Chang is approved:  \n\n|  |\n| --- |\n|  |\n|  |\n|  |\n\nCommittee Chairperson  \nUniversity of California, Riverside  \nABSTRACT OF THE DISSERTATION  \nMachine Learning in Resource-Constrained Distributed and Dynamic Environments  \nby  \nXiangyu Chang  \nDoctor of Philosophy, Graduate Program in Electrical Engineering University of California, Riverside, September 2025  \nProf. Amit K. Roy-Chowdhury, Chairperson  \nModern machine learning models, despite their power, face significant challenges related to efficiency and adaptability when deployed in distributed and dynamic real-world environments, especially under tight resource constraints. This dissertation presents novel algorithms and theoretical analyses designed to address these critical gaps across three distinct but related frontiers of machine learning.  \nFirst, we tackle the problem of adapting models on resource-constrained edge devices to unlabeled data from dynamic distributions. We introduce AdMiT (Adaptive Multi-Source Tuning), a framework that leverages a pre-trained library of Parameter-Efficient Tuning (PET) modules [30 , 31] . AdMiT dynamically selects and integrates a sparse subset of these modules by matching the target data distribution to source distributions using Kernel Mean Embedding (KME) [25] . This approach is highly efficient, privacy-preserving, and obviates the need for auxiliary routing networks, demonstrating superior adaptation performance in extensive benchmarks.  \nSecond, the dissertation addresses a fundamental challenge in Asynchronous Feder-  \nated Learning (AFL) [2], a phenomenon we term \"heterogeneity amplification\". This issue arises when the asynchronous nature of client updates interacts with non-IID data, causing faster clients to bias the global model [37] . To counter this, we propose the All-Client AFL (AC-AFL) algorithm, which mitigates this bias by aggregating the latest available gradient from all clients in every server update. Our theoretical framework, built upon a novel mean square error (MSE) decomposition, and empirical results confirm that AC-AFL achieves superior convergence and robustness.  \nFinally, we investigate the synergistic relationship between in-context learning (ICL) [10] and prompt-tuning in large language models. Focusing on a single-layer linear attention model, we introduce the theoretical concept of \"covariance-mean decoupling\" to formally explain how task-specific parameters collaborate. Our analysis reveals that prompts primarily learn to explain the conditional mean of a task distribution, while the attention mechanism explains its variance. This perspective provides a provable explanation for why jointly training prompts and attention weights is more effective than sequential fine-tuning.  \nCollectively, this body of work contributes practical algorithms and foundational theoretical insights that enhance the efficiency and adaptability of machine learning systems in complex, real-world deployments, particularly for resource","cbCaimX7xYvjh4RF","https://ap.wps.com/l/cbCaimX7xYvjh4RF","pdf",3185527,1,239,"English","en",105,"# 1 Introduction\n# 2 AdMiT: Adaptive Multi-Source Tuning in Dynamic Environments\n## 2.1 Introduction\n## 2.3 Proposed Method: AdMiT\n## 2.4 Evaluations\n# 3 Mitigating Heterogeneity Amplification in Asynchronous Federated Learning\n## 3.1 Introduction\n## 3.3 Preliminaries and Analytical Framework\n## 3.5 Experimental Results\n# 4 Provable Benefits of Task-Specific Prompts for In-context Learning\n## 4.1 Introduction\n## 4.3 Setup and Preliminaries\n## 4.4 Main Results","[{\"question\":\"What core deployment challenges does the dissertation address?\",\"answer\":\"It targets efficiency and adaptability problems when modern machine learning models run in distributed and dynamic real-world settings under tight resource constraints.\"},{\"question\":\"How does AdMiT adapt models on edge devices without labeled data?\",\"answer\":\"AdMiT adaptively selects and integrates parameter-efficient tuning modules by matching the target data distribution to source distributions using kernel mean embedding.\"},{\"question\":\"What is heterogeneity amplification in asynchronous federated learning, and how is it mitigated?\",\"answer\":\"Heterogeneity amplification arises when asynchronous client updates interact with non-IID data, causing faster clients to bias the global model. The AC-AFL algorithm mitigates this by aggregating the latest available gradients from all clients at each server update.\"},{\"question\":\"What theoretical insight does the dissertation provide for in-context learning with prompt-tuning?\",\"answer\":\"For a single-layer linear attention model, it introduces covariance-mean decoupling to explain how prompts and attention weights collaborate, showing that prompts primarily capture conditional means while attention explains variance.\"}]","Machine Learning in Resource-Constrained Distributed and Dynamic Environments - Dissertation | PDF",1785735987,602,{"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":90,"head_meta":92,"extra_data":94,"updated_unix":28},"machine-learning-in-resource-constrained-distributed-and-dynamic-environments-dissertation","",{"@graph":36,"@context":89},[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/machine-learning-in-resource-constrained-distributed-and-dynamic-environments-dissertation/121505/",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","2026-08-03",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81,85],{"name":72,"@type":73,"acceptedAnswer":74},"What core deployment challenges does the dissertation address?","Question",{"text":75,"@type":76},"It targets efficiency and adaptability problems when modern machine learning models run in distributed and dynamic real-world settings under tight resource constraints.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does AdMiT adapt models on edge devices without labeled data?",{"text":80,"@type":76},"AdMiT adaptively selects and integrates parameter-efficient tuning modules by matching the target data distribution to source distributions using kernel mean embedding.",{"name":82,"@type":73,"acceptedAnswer":83},"What is heterogeneity amplification in asynchronous federated learning, and how is it mitigated?",{"text":84,"@type":76},"Heterogeneity amplification arises when asynchronous client updates interact with non-IID data, causing faster clients to bias the global model. The AC-AFL algorithm mitigates this by aggregating the latest available gradients from all clients at each server update.",{"name":86,"@type":73,"acceptedAnswer":87},"What theoretical insight does the dissertation provide for in-context learning with prompt-tuning?",{"text":88,"@type":76},"For a single-layer linear attention model, it introduces covariance-mean decoupling to explain how prompts and attention weights collaborate, showing that prompts primarily capture conditional means while attention explains variance.","https://schema.org",{"og:url":52,"og:type":91,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":93,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":96},[97,101,105,109,114,119,124,127,132,135,139],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":106,"show_sort_weight":107,"slug":108},"Exam",70,"exam",{"id":110,"doc_module":4,"doc_module_name":46,"category_name":111,"show_sort_weight":112,"slug":113},5,"Comic",60,"comic",{"id":115,"doc_module":4,"doc_module_name":46,"category_name":116,"show_sort_weight":117,"slug":118},6,"Technology",50,"technology",{"id":120,"doc_module":4,"doc_module_name":46,"category_name":121,"show_sort_weight":122,"slug":123},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":125,"slug":126},30,"research-report",{"id":128,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":130,"slug":131},9,"Religion & Spirituality",20,"religion-spirituality",{"id":130,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":130,"slug":134},"World Cup","world-cup",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":136,"slug":138},10,"Lifestyle","lifestyle",{"id":140,"doc_module":4,"doc_module_name":46,"category_name":141,"show_sort_weight":110,"slug":142},19,"General","general"]