CINXE.COM
Adaptive Adversaries in Byzantine-Robust Federated Learning: A survey.
<!DOCTYPE html> <html lang="en"> <head> <meta charset="utf-8"> <meta name="viewport" content="width=device-width, initial-scale=1, shrink-to-fit=no"> <link href="/css/dist/css/bootstrap.min.css" rel="stylesheet"> <title>Adaptive Adversaries in Byzantine-Robust Federated Learning: A survey.</title> <link rel="stylesheet" href="/css/eprint.css?v=10"> <style> a.toggle-open:after { content:' -'; font-weight: 800; } a.toggle-closed:after { content: " ›"; font-weight: 800; } .paper-abstract { white-space: pre-wrap; } #metadata dt { margin-top: 1rem; } #metadata dt + dd { /* gap between dt and first dd */ margin-top: .75rem; } #metadata dd { margin-left: 2rem; } #metadata dd.keywords { padding-bottom: .5rem; } span.authorName { margin-top: .5rem; font-style: italic; } </style> <script> MathJax = { tex: { inlineMath: [['$', '$'], ['\\(', '\\)']], displayMath: [ ['$$','$$'], ["\\[","\\]"] ], processEnvironments: false }, loader: { load: [ "ui/safe", "ui/lazy", ], }, options: { safeOptions: { allow: { URLs: "none", classes: "safe", cssIDs: "safe", styles: "safe", }, }, } }; </script> <script id="MathJax-script" async src="/js/mathjax/tex-chtml.js"></script> <meta name="citation_title" content="Adaptive Adversaries in Byzantine-Robust Federated Learning: A survey."> <meta name="citation_author" content="Jakub Kacper Szeląg"> <meta name="citation_author" content="Ji-Jian Chin"> <meta name="citation_author" content="Sook-Chin Yip"> <meta name="citation_journal_title" content="Cryptology ePrint Archive"> <meta name="citation_publication_date" content="2025"> <meta name="citation_pdf_url" content="https://eprint.iacr.org/2025/510.pdf"> <meta property="og:image" content="https://eprint.iacr.org/img/iacrlogo.png"/> <meta property="og:image:alt" content="IACR logo"/> <meta property="og:url" content="https://eprint.iacr.org/2025/510"> <meta property="og:site_name" content="IACR Cryptology ePrint Archive" /> <meta property="og:type" content="article" /> <meta property="og:title" content="Adaptive Adversaries in Byzantine-Robust Federated Learning: A survey." /> <meta property="og:description" content="Federated Learning (FL) has recently emerged as one of the leading paradigms for collaborative machine learning, serving as a tool for model computation without a need to expose one’s privately stored data. However, despite its advantages, FL systems face severe challenges within its own security solutions that address both privacy and robustness of models. This paper focuses on vulnerabilities within the domain of FL security with emphasis on model-robustness. Identifying critical gaps in current defences, particularly against adaptive adversaries which modify their attack strategies after being disconnected and rejoin systems to continue attacks. To our knowledge, other surveys in this domain do not cover the concept of adaptive adversaries, this along with the significance of their impact serves as the main motivation for this work. Our contributions are fivefold: (1) we present a comprehensive overview of FL systems, presenting a complete summary of its fundamental building blocks, (2) an extensive overview of existing vulnerabilities that target FL systems in general, (3) highlight baseline attack vectors as well as state-of-the-art approaches to development of attack methods and defence mechanisms, (4) introduces a novel baseline method of attack leveraging reconnecting malicious clients, and (5) identifies future research directions to address and counter adaptive attacks. We demonstrate through experimental results that existing baseline secure aggregation rules used in other works for comparison such as Krum and Trimmed Mean are insufficient against those attacks. Further, works improving upon those algorithms do not address this concern either. Our findings serve as a basis for redefining FL security paradigms in the direction of adaptive adversaries." /> <meta property="article:section" content="PROTOCOLS" /> <meta property="article:modified_time" content="2025-03-21T05:23:02+00:00" /> <meta property="article:published_time" content="2025-03-18T15:44:42+00:00" /> <meta property="article:tag" content="Machine Learning" /> <meta property="article:tag" content="Federated Learning" /> <meta property="article:tag" content="Secure Aggregation" /> <meta property="article:tag" content="Adaptive Adversaries" /> <meta property="article:tag" content="Byzantine-Robust Aggregation" /> </head> <body> <noscript> <h1 class="text-center">What a lovely hat</h1> <h4 class="text-center">Is it made out of <a href="https://iacr.org/tinfoil.html">tin foil</a>?</h4> </noscript> <div class="fixed-top" id="topNavbar"> <nav class="navbar navbar-custom navbar-expand-lg"> <div class="container px-0 justify-content-between justify-content-lg-evenly"> <div class="order-0 align-items-center d-flex"> <button class="navbar-toggler btnNoOutline" type="button" data-bs-toggle="collapse" data-bs-target="#navbarContent" aria-controls="navbarContent" aria-expanded="false"> <span class="icon-bar top-bar"></span> <span class="icon-bar middle-bar"></span> <span class="icon-bar bottom-bar"></span> </button> <a class="d-none me-5 d-lg-inline" href="https://iacr.org/"><img class="iacrlogo" src="/img/iacrlogo_small.png" alt="IACR Logo" style="max-width:6rem;"></a> </div> <a class="ePrintname order-1" href="/"> <span class="longNavName">Cryptology ePrint Archive</span> </a> <div class="collapse navbar-collapse order-3" id="navbarContent"> <ul class="navbar-nav me-auto ms-2 mb-2 mb-lg-0 justify-content-end w-100"> <li class="ps-md-3 nav-item dropdown"> <a class="nav-link dropdown-toggle" href="#" id="navbarDropdown" role="button" data-bs-toggle="dropdown" aria-expanded="false"> Papers </a> <ul class="dropdown-menu me-3" aria-labelledby="navbarDropdown"> <span class="text-dark mx-3" style="white-space:nowrap;">Updates from the last:</span> <li><a class="dropdown-item ps-custom" href="/days/7">7 days</a></li> <li><a class="dropdown-item ps-custom" href="/days/31">31 days</a></li> <li><a class="dropdown-item ps-custom" href="/days/183">6 months</a></li> <li><a class="dropdown-item ps-custom" href="/days/365">365 days</a></li> <li><hr class="dropdown-divider"></li> <li><a class="dropdown-item" href="/byyear">Listing by year</a></li> <li><a class="dropdown-item" href="/complete">All papers</a></li> <li><a class="dropdown-item" href="/complete/compact">Compact view</a></li> <li><a class="dropdown-item" href="https://www.iacr.org/news/subscribe">Subscribe</a></li> <li><hr class="dropdown-divider"></li> <li><a class="dropdown-item" href="/citation.html">How to cite</a></li> <li><hr class="dropdown-divider"></li> <li><a class="dropdown-item" href="/rss">Harvesting metadata</a></li> </ul> </li> <li class="ps-md-3 nav-item dropdown"> <a class="nav-link dropdown-toggle" href="#" id="submissionsDropdown" role="button" data-bs-toggle="dropdown" aria-expanded="false"> Submissions </a> <ul class="dropdown-menu me-3" aria-labelledby="submissionsDropdown"> <li><a class="dropdown-item" href="/submit">Submit a paper</a></li> <li><a class="dropdown-item" href="/revise">Revise or withdraw a paper</a></li> <li><a class="dropdown-item" href="/operations.html">Acceptance and publishing conditions</a></li> </ul> </li> <li class="ps-md-3 nav-item dropdown"> <a class="nav-link dropdown-toggle" href="#" id="aboutDropdown" role="button" data-bs-toggle="dropdown" aria-expanded="false"> About </a> <ul class="dropdown-menu me-3" aria-labelledby="aboutDropdown"> <li><a class="dropdown-item" href="/about.html">Goals and history</a></li> <li><a class="dropdown-item" href="/news.html">News</a></li> <li><a class="dropdown-item" href="/stats">Statistics</a></li> <li><a class="dropdown-item" href="/contact.html">Contact</a></li> </ul> </li> </ul> </div> <div class="dropdown ps-md-2 text-right order-2 order-lg-last"> <button class="btn btnNoOutline" type="button" id="dropdownMenuButton1" data-bs-toggle="dropdown" aria-expanded="false"> <img src="/img/search.svg" class="searchIcon" alt="Search Button"/> </button> <div id="searchDd" class="dropdown-menu dropdown-menu-end p-0" aria-labelledby="dropdownMenuButton1"> <form action="/search" method="GET"> <div class="input-group"> <input id="searchbox" name="q" type="search" class="form-control" autocomplete="off"> <button class="btn btn-secondary border input-group-append ml-2"> Search </button> </div> </form> <div class="ms-2 p-1 d-none"><a href="/search">Advanced search</a></div> </div> </div> </div> </nav> </div> <main id="eprintContent" class="container px-3 py-4 p-md-4"> <div class="row mt-4"> <div class="col-md-7 col-lg-8 pe-md-5"> <h4>Paper 2025/510</h4> <h3 class="mb-3">Adaptive Adversaries in Byzantine-Robust Federated Learning: A survey.</h3> <div class="author"><span class="authorName">Jakub Kacper Szeląg</span><a class="ms-1" target="_blank" href="https://orcid.org/0009-0000-9635-0598"><img class="align-baseline orcidIcon" src="/img/orcid.svg"></a><span class="affiliation">, University of Plymouth</span></div> <div class="author"><span class="authorName">Ji-Jian Chin</span><a class="ms-1" target="_blank" href="https://orcid.org/0000-0001-9809-6976"><img class="align-baseline orcidIcon" src="/img/orcid.svg"></a><span class="affiliation">, University of Plymouth</span></div> <div class="author"><span class="authorName">Sook-Chin Yip</span><a class="ms-1" target="_blank" href="https://orcid.org/0000-0002-8575-4244"><img class="align-baseline orcidIcon" src="/img/orcid.svg"></a><span class="affiliation">, Multimedia University</span></div> <h5 class="mt-3">Abstract</h5> <p style="white-space: pre-wrap;">Federated Learning (FL) has recently emerged as one of the leading paradigms for collaborative machine learning, serving as a tool for model computation without a need to expose one’s privately stored data. However, despite its advantages, FL systems face severe challenges within its own security solutions that address both privacy and robustness of models. This paper focuses on vulnerabilities within the domain of FL security with emphasis on model-robustness. Identifying critical gaps in current defences, particularly against adaptive adversaries which modify their attack strategies after being disconnected and rejoin systems to continue attacks. To our knowledge, other surveys in this domain do not cover the concept of adaptive adversaries, this along with the significance of their impact serves as the main motivation for this work. Our contributions are fivefold: (1) we present a comprehensive overview of FL systems, presenting a complete summary of its fundamental building blocks, (2) an extensive overview of existing vulnerabilities that target FL systems in general, (3) highlight baseline attack vectors as well as state-of-the-art approaches to development of attack methods and defence mechanisms, (4) introduces a novel baseline method of attack leveraging reconnecting malicious clients, and (5) identifies future research directions to address and counter adaptive attacks. We demonstrate through experimental results that existing baseline secure aggregation rules used in other works for comparison such as Krum and Trimmed Mean are insufficient against those attacks. Further, works improving upon those algorithms do not address this concern either. Our findings serve as a basis for redefining FL security paradigms in the direction of adaptive adversaries.</p> </div> <div id="metadata" class="col-md-5 col-lg-4 ps-md-5 mt-4 mt-md-0"> <h5>Metadata</h5> <dl> <dt> Available format(s) </dt> <dd> <a class="btn btn-sm btn-outline-dark" href="/2025/510.pdf"> <img class="icon" src="/img/file-pdf.svg">PDF</a> </dd> <dt>Category</dt> <dd><a href="/search?category=PROTOCOLS"><small class="badge category category-PROTOCOLS">Cryptographic protocols</small></a></dd> <dt>Publication info</dt> <dd>Preprint. </dd> <dt>Keywords</dt> <dd class="keywords"><a href="/search?q=Machine%20Learning" class="me-2 badge bg-secondary keyword">Machine Learning</a><a href="/search?q=Federated%20Learning" class="me-2 badge bg-secondary keyword">Federated Learning</a><a href="/search?q=Secure%20Aggregation" class="me-2 badge bg-secondary keyword">Secure Aggregation</a><a href="/search?q=Adaptive%20Adversaries" class="me-2 badge bg-secondary keyword">Adaptive Adversaries</a><a href="/search?q=Byzantine-Robust%20Aggregation" class="me-2 badge bg-secondary keyword">Byzantine-Robust Aggregation</a></dd> <dt>Contact author(s)</dt> <dd><span class="font-monospace"> jakub szelag<span class="obfuscate"> @ </span>students plymouth ac uk<br>ji-jian chin<span class="obfuscate"> @ </span>plymouth ac uk<br>scyip<span class="obfuscate"> @ </span>mmu edu my </span> </dd> <dt>History</dt> <dd>2025-03-21: revised</dd> <dd>2025-03-18: received</dd> <dd><a rel="nofollow" href="/archive/versions/2025/510">See all versions</a></dd> <dt>Short URL</dt> <dd><a href="https://ia.cr/2025/510">https://ia.cr/2025/510</a></dd> <dt>License</dt> <dd><a rel="license" target="_blank" href="https://creativecommons.org/licenses/by/4.0/"> <img class="licenseImg" src="/img/license/CC_BY.svg" alt="Creative Commons Attribution" title="Creative Commons Attribution"><br> <small>CC BY</small> </a> </dd> </dl> </div> </div> <p class="mt-4"><strong>BibTeX</strong> <button id="bibcopy" class="ms-2 btn btn-sm btn-outline-dark" aria-label="Copy to clipboard" onclick="copyBibtex()"> <img src="/img/copy-outline.svg" class="icon">Copy to clipboard</button></p> <pre id="bibtex"> @misc{cryptoeprint:2025/510, author = {Jakub Kacper Szeląg and Ji-Jian Chin and Sook-Chin Yip}, title = {Adaptive Adversaries in Byzantine-Robust Federated Learning: A survey.}, howpublished = {Cryptology {ePrint} Archive, Paper 2025/510}, year = {2025}, url = {https://eprint.iacr.org/2025/510} } </pre> <script> var bibcopy; function triggerTooltip() { console.log('setting tooltip'); } window.onload = triggerTooltip; function copyBibtex() { let range = document.createRange(); range.selectNode(document.getElementById('bibtex')); window.getSelection().removeAllRanges(); window.getSelection().addRange(range); document.execCommand('copy'); window.getSelection().removeAllRanges(); let bibcopy = document.getElementById('bibcopy'); let copyTooltip = new bootstrap.Tooltip(bibcopy, {trigger: 'manual', title: 'Copied!'}); copyTooltip.show(); setTimeout(function() { copyTooltip.dispose(); }, 2000); } </script> </main> <div class="container-fluid mt-auto" id="eprintFooter"> <a href="https://iacr.org/"> <img id="iacrlogo" src="/img/iacrlogo_small.png" class="img-fluid d-block mx-auto" alt="IACR Logo"> </a> <div class="colorDiv"></div> <div class="alert alert-success w-75 mx-auto"> Note: In order to protect the privacy of readers, eprint.iacr.org does not use cookies or embedded third party content. </div> </div> <script src="/css/bootstrap/js/bootstrap.bundle.min.js"></script> <script> var topNavbar = document.getElementById('topNavbar'); if (topNavbar) { document.addEventListener('scroll', function(e) { if (window.scrollY > 100) { topNavbar.classList.add('scrolled'); } else { topNavbar.classList.remove('scrolled'); } }) } </script> </body> </html>