[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-291223-105":59,"doc-detail-291223-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","privacy-preserving-machine-learning-on-graphs-thesis-abstract","Privacy-Preserving Machine Learning on Graphs - Thesis Abstract","","Graph Neural Networks (GNNs) enable powerful learning on graph-structured data, yet they raise urgent privacy risks because graphs often contain sensitive personal information such as social interactions and financial activities. This thesis examines privacy-preserving GNNs using Differential Privacy (DP) to provide formal privacy guarantees. It develops locally differentially private GCN methods for privately collecting node features and estimating convolution, including perturbation, compression, denoising, and robust learning under noisy labels. It further proposes global DP training via aggregation perturbation (AP) and progressive private embedding-based models, achieving strong privacy-utility trade-offs and broad application potential.",{"@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/privacy-preserving-machine-learning-on-graphs-thesis-abstract/291223/",{"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/privacy-preserving-machine-learning-on-graphs-thesis-abstract/291223.png","ImageObject",300,407,{"name":92,"@type":93},"Ophelia","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",14,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"Why does learning on graphs require privacy protection?","Question",{"text":113,"@type":114},"Graphs frequently contain sensitive personal information such as social interactions and financial transactions. Without protection, training or inference can leak private data.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"What Differential Privacy framework does the thesis use?",{"text":118,"@type":114},"The thesis focuses on Differential Privacy (DP) as a rigorous mathematical framework to formalize and guarantee privacy for machine learning algorithms.",{"name":120,"@type":111,"acceptedAnswer":121},"What is the main contribution for global DP training of GNNs?",{"text":122,"@type":114},"The thesis introduces aggregation perturbation (AP), which injects noise into the neighborhood aggregation step to enable private learning and inference with DP guarantees.","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},291223,1789643791,{"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},7971461741311,"https://ap-avatar.wpscdn.com/avatar/74000253aff267980c6?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779345379180704826","Thèse n° 10 245  \nPrivacy-Preserving Machine Learning on Graphs  \nPrésentée le 1er décembre 2023  \nFaculté des sciences et techniques de l’ingénieur Laboratoire de l’IDIAP  \nProgramme doctoral en génie électrique  \npour l’obtention du grade de Docteur ès Sciences  \npar  \nSina SAJADMANESH  \nAcceptée sur proposition du jury  \nDr D. Gillet, président du jury  \nProf. D. Gatica-Perez, directeur de thèse Prof. H. Haddadi, rapporteur  \nProf. E. De Cristofaro, rapporteur Prof. K. Aberer, rapporteur  \n2023  \nThe roots of education are bitter, but the fruit is sweet.  \n—Aristotle  \nTo my dear parents and my loving wife, whose unwavering affection, support, and encouragement throughout this academic journey illuminated my path. . .  \nAcknowledgements  \nI would like to express my heartfelt gratitude to the following individuals and institutions who have been instrumental in the completion of my PhD thesis. First, I am deeply indebted tomy advisor, Prof. Daniel Gatica-Perez, for his unwavering support, patience, and invaluable insights throughout this journey. Second, I extend my sincere appreciation to the members of my examining committee: Dr. Denis Gillet, Prof. Karl Aberer, Prof. Emiliano De Cristofaro, and Prof. Hamed Haddadi. Their constructive feedback and support were instrumental in shaping this research. Third, I acknowledge the financial support provided by the Swiss National Science Foundation (Dusk2Dawn Project) and the European Commission (AI4Media Project) . Their generous support allowed me to conduct this research and push the boundaries of knowledge in my field. Fourth, I am deeply grateful to my fellow researchers and collaborators, both within and outside my institution, for their camaraderie, brainstorming sessions, and contributions that enriched the research experience. Fifth, I would like to thank all my friends and colleagues at Idiap Research Institute and EPFL who stood by me, offering motivation, support, laughter, and sometimes a much-needed break from academic pursuits. Last but not least, I express my deepest gratitude to my family for their unwavering love and support throughout my life. Their sacrifices and understanding made this journey possible.  \nThis thesis represents the culmination of years of hard work and dedication from many individuals and organizations. I am humbled by the collective support and encouragement that has propelled me forward in my academic pursuits.  \nThankyou all for your unwavering belief in me and for being an integral part of this academic journey.  \nLausanne, September 19, 2023 Sina Sajadmanesh  \nAbstract  \nGraph Neural Networks (GNNs) have emerged as a powerful tool for learning on graphs, demonstrating exceptional performance in various domains. However, as GNNs become increasingly popular, new challenges arise. One of the most pressing is the need to ensure privacy when learning from graphs since they often contain personal information like social interactions and financial transactions. Balancing the preservation of privacy with the effectiveness ofGNNs is a crucial yet challenging task due to the unique relational nature of graphs, which sets them apart from other data types like images and text. This thesis presents an in-depth exploration of privacy-preserving GNNs, explicitly focusing on Differential Privacy (DP) as a rigorous mathematical framework for formalizing the privacy guarantees of machine learning algorithms.  \nFirst, we focus on local DP and propose a privacy-preserving Graph Convolutional Network (GCN) model, allowing a central server to efficiently communicate with the graph nodes to privately collect their features and estimate the first-layer graph convolution of the GNN. Building on the same foundations, we then propose an architecture-agnostic approach for settings where all the node-specific data, including features and labels, are private and locally stored on the nodes, but the edges remain public to the central server that trains the GNN. Our contr","cbCaihHo3Xyg9let","https://ap.wps.com/l/cbCaihHo3Xyg9let","pdf",4162340,165,"English","# Abstract\n## Local Differential Privacy on Graphs\n## Global Differential Privacy and Aggregation Perturbation\n## Empirical Evaluation and Applications","[{\"question\":\"Why does learning on graphs require privacy protection?\",\"answer\":\"Graphs frequently contain sensitive personal information such as social interactions and financial transactions. Without protection, training or inference can leak private data.\"},{\"question\":\"What Differential Privacy framework does the thesis use?\",\"answer\":\"The thesis focuses on Differential Privacy (DP) as a rigorous mathematical framework to formalize and guarantee privacy for machine learning algorithms.\"},{\"question\":\"What is the main contribution for global DP training of GNNs?\",\"answer\":\"The thesis introduces aggregation perturbation (AP), which injects noise into the neighborhood aggregation step to enable private learning and inference with DP guarantees.\"}]","Privacy-Preserving Machine Learning on Graphs - Thesis Abstract | PDF",416]