[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-290793-105":59,"doc-detail-290793-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","on-the-privacy-robustness-utility-trilemma-in-distributed-learning-abstract","On the Privacy-Robustness-Utility Trilemma in Distributed Learning - Abstract","","Distributed machine learning in sensitive applications requires algorithms that protect privacy, remain robust to faults and adversarial behavior, and still deliver useful performance. Prior work treats privacy and robustness separately, leaving their combined behavior unclear. This study provides a tight error analysis for algorithms that ensure robustness against adversarial machines and differential privacy for honest workers, proving a fundamental privacy–robustness–utility trade-off, with matching lower and upper bounds.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/on-the-privacy-robustness-utility-trilemma-in-distributed-learning-abstract/290793/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/on-the-privacy-robustness-utility-trilemma-in-distributed-learning-abstract/290793.png","ImageObject",300,407,{"name":92,"@type":93},"River Wang","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-29","2026-09-17",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",12,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does the paper address in distributed learning?","Question",{"text":113,"@type":114},"It studies the combined challenge of ensuring data privacy, robustness to adversarial/faulty workers, and maintaining utility in distributed machine learning.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"What main result does the paper present?",{"text":118,"@type":114},"It provides a tight analysis showing a fundamental trade-off among privacy, robustness, and utility, including matching lower and upper bounds.",{"name":120,"@type":111,"acceptedAnswer":121},"How does the paper prove the lower and upper bounds?",{"text":122,"@type":114},"For the lower bound, it analyzes mean estimation under distributed differential privacy and robustness constraints via reductions to centralized one-way marginal estimation. For the upper bound, it proposes a distributed learning algorithm using a high-dimensional robust aggregation rule.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},290793,1789642577,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":130,"read_time":145,"preview_page_count":4},1099514067438,"https://ap-avatar.wpscdn.com/avatar/100002539ee87300030?x-image-process=image/resize,m_fixed,w_180,h_180&k=1780474512215547542","On the Privacy-Robustness-Utility Trilemma in Distributed Learning  \nYoussef Allouah 1 Rachid Guerraoui 1 Nirupam Gupta 1 Rafal Pinot 1 John Stephan 1  \nAbstract  \nThe ubiquity of distributed machine learning (ML) in sensitive public domain applications calls for algorithms that protect data privacy, while being robust to faults and adversarial behaviors. Although privacy and robustness have been extensively studied independently in distributed ML, their synthesis remains poorly understood. We present the first tight analysis of the error incurred by any algorithm ensuring robustness against a fraction of adversarial machines, as well as differential privacy (DP) for honest machines’ data against any other curious entity. Our analysis exhibits a fundamental trade-off between privacy, robustness, and utility. To prove our lower bound, we consider the case of mean estimation, subject to distributed DP and robustness constraints, and devise reductions to centralized estimation of one-way marginals. We prove our matching upper bound by presenting a new distributed ML algorithm using a high-dimensional robust aggregation rule. The latter amortizes the dependence on the dimension in the error (caused by adversarial workers and DP), while being agnostic to the statistical properties of the data.  \n1. Introduction  \nDistributed machine learning (ML) has been playing a pivotal role in a wide range of applications (Dean et al., 2012; Abadi et al., 2016), due to an unprecedented growth in the complexity of ML models and the volume of data being used for training purposes. Distributed ML breaks a complex ML task into sub-tasks that are performed in a collaborative fashion. In the standard server-based architecture, n machines (a.k.a., workers) collaboratively train a global model on their datasets, with the help of a coordinator (the server) . This is typically achieved through a distributed implementation  \n1Ecole Polytechnique Fdrale de Lausanne (EPFL),  \nSwitzerland. Correspondence to: Youssef Allouah  \n\u003C[youssef.allouah@epfl.ch](youssef.allouah@epfl.ch) >.  \nProceedings of the 40 th International Conference on Machine Learning, Honolulu, Hawaii, USA. PMLR 202, 2023 . Copyright 2023 by the author(s) .  \nof the renowned stochastic gradient descent (SGD) algorithm (Bertsekas & Tsitsiklis, 2015) . In distributed SGD (or DSGD), the server maintains a model which is updated iteratively by averaging gradients of the loss function associated with the model, computed by the different workers upon sampling random points from their local datasets. DSGD is particularly useful in cases where the data held by the workers is too sensitive to be shared, e.g., medical data collected by several hospitals (Sheller et al., 2020) .  \nPrivacy. Although DSGD inherently ensures privacy of the workers’ data to an extent, by not sharing it explicitly, information leakage can still be significant. When the ML model maintained at the server is publicly released, it maybe exposed to membership inference (Shokri et al., 2016) or model inversion attacks (Fredrikson et al., 2015; Hitaj et al., 2017; Melis et al., 2019) by external entities. Furthermore, upon observing the gradients and transient models during the learning procedure, curious machines (be they workers or the server itself) can infer sensitive information about the datasets held locally by the machines, or even reconstruct data points in certain scenarios (Phong et al., 2017; Wanget al., 2019b; Zhu et al., 2019; Zhao et al., 2020) .  \nRobustness. In real-world distributed systems, it is arguably inevitable to encounter faulty workers that may deviate from their prescribed algorithm. This may result from hardware and software bugs, data corruption, network latency, or malicious adversaries controlling a subset of workers. To cover all such possible scenarios, it is common to assume that a fraction of the machines can be adversarial1 and arbitrarily deviate from their algorithms. In the context of DSGD, ","cbCailmJdy5LAcOj","https://ap.wps.com/l/cbCailmJdy5LAcOj","pdf",1284716,58,"English","# Introduction\n## Privacy\n## Robustness\n## Integrating privacy and robustness","[{\"question\":\"What problem does the paper address in distributed learning?\",\"answer\":\"It studies the combined challenge of ensuring data privacy, robustness to adversarial/faulty workers, and maintaining utility in distributed machine learning.\"},{\"question\":\"What main result does the paper present?\",\"answer\":\"It provides a tight analysis showing a fundamental trade-off among privacy, robustness, and utility, including matching lower and upper bounds.\"},{\"question\":\"How does the paper prove the lower and upper bounds?\",\"answer\":\"For the lower bound, it analyzes mean estimation under distributed differential privacy and robustness constraints via reductions to centralized one-way marginal estimation. For the upper bound, it proposes a distributed learning algorithm using a high-dimensional robust aggregation rule.\"}]","On the Privacy-Robustness-Utility Trilemma in Distributed Learning - Abstract | PDF",146]