[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86254-en":3,"doc-seo-86254-105":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},86254,687197207919,"Theodora","https://ap-avatar.wpscdn.com/avatar/a000253d6f5f7c60be?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779446848396160552",8,"Research & Report","HermesHFL Incentive-Compatible Hierarchical Federated Unlearning for Dynamic LLM Fine-Tuning","Hierarchical federated unlearning (HFUL) for large language model fine-tuning faces difficulties from hierarchical aggregation dependencies, dynamic client participation, and strong parameter coupling. Selectively removing the influence of particular participating clients becomes harder when unlearning requests overlap with client departures and later rejoining. HermesHFL proposes an incentive-compatible hierarchical rejoinable federated learning framework with selective unlearning. It supports dynamic participation and reintegration using PEFT with LoRA, formulates a unified bilevel optimization, and employs Neogen for efficient incentive and unlearning decision learning. Experiments confirm improved utility, unlearning effectiveness, convergence stability, and resource efficiency.","HermesHFL: Incentive-Compatible Hierarchical Federated Unlearning for Dynamic LLM Fine-Tuning  \nChenxi Sun, Student Member, IEEE , Minghui Liwang, Senior Member, IEEE, Wusi He, Zhang Liu, Yuhan Su, Seyyedali Hosseinalipour, Senior Member, IEEE , Xianbin Wang, Fellow, IEEE, Yiguang Hong, Fellow,  \nIEEE  \n~~ ~~ ✦ ~~ ~~  \narXiv :2607 . 1 1528v 1 [ cs .CE] 13 Jul 2026  \nAbstract—Hierarchical federated unlearning (HFUL) for large language model (LLM) fine-tuning introduces fundamental challenges stemming from hierarchical aggregation dependencies, dynamic client participation, and the strong parameter coupling inherent to LLM adaptation. In particular, selectively removing the influence of participating clients becomes substantially more difficult in HFUL, where model updates propagate through multiple aggregation stages and client unlearning requests may occur concurrently with client departures and subsequent rejoining behaviors. To address these challenges, we propose HermesHFL (hierarchical rejoinable machine learning with selective unlearning for incentive-compatible HFL), a hierarchical rejoinable federated learning framework with selective unlearning for incentive-compatible LLM finetuning. HermesHFL supports dynamic client participation, selective unlearning, and client reintegration while enabling scalable LLM adaptation through parameter-efficient fine-tuning (PEFT) with LoRA. We formulatea unified optimization problem that jointly captures client participation, edge association, incentive allocation, and unlearning decisions under heterogeneous and strategic client behaviors. To efficiently solve this challenging problem, we develop Neogen (neural network guided dual network evolutionary optimization), a neural-guided bilevel evolutionary optimization framework that combines covariance matrix adaptation evolution strategy (CMA-ES) for continuous incentive optimization with cross-generational elitist selection, heterogeneous recombination, and cataclysmic mutation evolutionary algorithm (CHC)-based mechanism for discrete client participation and association decisions. A neural surrogate guidance mechanism is further introduced to accelerate convergence and reduce search complexity. Extensive experiments on LLMfine-tuning tasks demonstrate that HermesHFL consistently outperforms state-of-the-art baselines in terms of model utility, unlearning effectiveness, convergence stability, and resource efficiency.  \nIndex Terms—Hierarchical Federated Learning, Machine Unlearning, Large Language Models, LoRA, Incentive Mechanism, Dynamic Clients  \n1 INTRODUCTION  \nC. Sun (aloys [sun@outlook.com](sun@outlook.com)) is with the School of Economics and Management, Tongji University, Shanghai, China. M. Liwang (minghuili[wang@tongji.edu.cn](wang@tongji.edu.cn)) is with the Shanghai Research Institute for Intelligent Autonomous Systems, State Key Laboratory of Autonomous Intelligent Unmanned Systems, Frontiers Science Center for Intelligent Autonomous Systems, and Department of Control Science and Engineering, Tongji University, Shanghai, China. S. Hosseinalipour ([alipour@buffalo.edu](alipour@buffalo.edu)) is with Department of Electrical Engineering, University at Buffalo–SUNY, NY, USA.  \nTHE proliferation of smart devices has led to mas  \nsive data generation across the Internet-of-Things (IoT) ecosystem, creating unprecedented opportunities for machine learning (ML) . However, increasingly stringent privacy regulations and data-governance constraints make conventional centralized ML pipelines impractical. This challenge has driven the emergence of distributed learning paradigms, among which federated learning (FL) [1] stands out as a representative framework for privacy-preserving collaborative model training. Despite its advantages, conventional FL suffers from heavy communication overhead, aggregation latency, and unstable convergence in largescale, heterogeneous networks. To address these limitations, hierarchical federated learning","cbCaigfWQsefTadz","https://ap.wps.com/l/cbCaigfWQsefTadz","pdf",4206423,2,1,17,"English","en",105,"# Abstract\n# Introduction\n## Core Motivation","[{\"question\":\"What main challenge does HermesHFL address in hierarchical federated unlearning for LLM fine-tuning?\",\"answer\":\"It targets the difficulty of selectively removing specific clients’ influence when multi-stage hierarchical aggregation propagates updates and unlearning requests can occur during dynamic client departure and rejoining.\"},{\"question\":\"How does HermesHFL support dynamic client participation and reintegration?\",\"answer\":\"It uses a hierarchical rejoinable federated learning design that enables clients to join, leave, and rejoin while still allowing selective unlearning under incentive-compatible incentives.\"},{\"question\":\"How are LLM fine-tuning and optimization made scalable in HermesHFL?\",\"answer\":\"HermesHFL applies parameter-efficient fine-tuning with LoRA for efficient adaptation, and uses Neogen, a neural-guided bilevel evolutionary optimization combining CMA-ES for continuous incentives with evolutionary mechanisms for discrete participation decisions, aided by a neural surrogate for faster convergence.\"}]",1784209842,43,{"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},"hermeshfl-incentive-compatible-hierarchical-federated-unlearning-for-dynamic-llm-fine-tuning","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/hermeshfl-incentive-compatible-hierarchical-federated-unlearning-for-dynamic-llm-fine-tuning/86254/",4,{"url":51,"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","2026-07-26","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 main challenge does HermesHFL address in hierarchical federated unlearning for LLM fine-tuning?","Question",{"text":75,"@type":76},"It targets the difficulty of selectively removing specific clients’ influence when multi-stage hierarchical aggregation propagates updates and unlearning requests can occur during dynamic client departure and rejoining.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does HermesHFL support dynamic client participation and reintegration?",{"text":80,"@type":76},"It uses a hierarchical rejoinable federated learning design that enables clients to join, leave, and rejoin while still allowing selective unlearning under incentive-compatible incentives.",{"name":82,"@type":73,"acceptedAnswer":83},"How are LLM fine-tuning and optimization made scalable in HermesHFL?",{"text":84,"@type":76},"HermesHFL applies parameter-efficient fine-tuning with LoRA for efficient adaptation, and uses Neogen, a neural-guided bilevel evolutionary optimization combining CMA-ES for continuous incentives with evolutionary mechanisms for discrete participation decisions, aided by a neural surrogate for faster convergence.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"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":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"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":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]