[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-119849-en":3,"doc-seo-119849-105":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":4,"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},119849,8796095360427,"Lucas Martin","https://ap-avatar.wpscdn.com/davatar_994ba38a5ba835b3df7d355c54d3ed8d",8,"Research & Report","On the Privacy Risks of Machine Learning Models - Dissertation","Machine learning (ML) advances of the past decade introduce escalating privacy risks as ML models are adopted in high-impact settings. These risks split into model vulnerability, where sensitive information can leak, and model abuse, where privacy can be violated by adversarial use. The dissertation studies both angles via membership inference attacks (MIA), analyzing label-only membership leakage and multi-exit networks, then addresses deepfake face manipulation through a defense that disrupts GAN inversion.","Saarland University  \nDepartment of Computer Science  \nOn the Privacy Risks of Machine Learning Models  \nDissertation  \nzur Erlangung des Grades  \ndes Doktors der Ingenieurwissenschaften  \nder Fakultät für Mathematik und Informatik  \nder Universität des Saarlandes  \nvon  \nZheng Li  \nSaarbrücken, 2023  \nTag des Kolloquiums: 05 Oktober 2023  \nDekan: Prof. Dr. Jürgen Steimle  \nPrüfungsausschuss:  \nVorsitzender: Pro. Dr. Ingmar Weber  \nBerichterstattende: Dr. Yang Zhang  \nProf. Dr. Mario Fritz  \nProf. Dr. Konrad Rieck  \nAkademischer Mitarbeiter: Dr. Zhiqiu Jiang  \nZusammenfassung  \nDas maschinelle Lernen (ML) hat in den letzten zehn Jahren enorme Fortschritte gemacht und wurde für eine breite Palette wichtiger Anwendungen eingesetzt. Durch den zunehmenden Einsatz von Modellen des maschinellen Lernens ist die Bedeutung von Datenschutzrisiken jedoch wichtiger denn je geworden. Diese Risiken können jenach der Rolle, die ML-Modelle spielen, in zwei Kategorien eingeteilt werden: in eine, in der die Modelle selbst anfällig für das Durchsickern sensibler Informationen sind, und in die andere, in der die Modelle zur Verletzung der Privatsphäre missbraucht werden.  \nIn dieser Dissertation untersuchen wir die Datenschutzrisiken von Modellen des maschinellen Lernens aus zwei Blickwinkeln, nämlich der Anfälligkeit von ML-Modellen und dem Missbrauch von ML-Modellen. Um die Anfälligkeit von ML-Modellen für Datenschutzrisiken zu untersuchen, führen wir zwei Studien zu einem der schwerwiegendsten Angriffe auf den Datenschutz von ML-Modellen durch, nämlich dem Angriff auf die Mitgliedschaft (membership inference attack, MIA) . Erstens erforschen wir das Durchsickern von Mitgliedschaften in ML-Modellen, die sich nur auf Labels beziehen. Wir präsentieren den ersten \"label-only membership inference\"-Angriff und stellen fest, dass das \"membership leakage\" schwerwiegender ist als bisher gezeigt. Zweitens führen wir die erste Analyse der Privatsphäre von Netzwerken mit mehreren Ausgängen durch die Linse des Mitgliedschaftsverlustes durch. Wir nutzen bestehende Angriffsmethoden, um die Anfälligkeit von Multi-Exit-Netzwerken für Membership-Inference-Angriffe zu quantifizieren und schlagen einen hybriden Angriff vor, der die Exit-Informationen ausnutzt, um die Angriffsleistung zu verbessern. Unter dem Gesichtspunkt des Missbrauchs von ML-Modellen zur Verletzung der Privatsphäre konzentrieren wir uns auf die Manipulation von Gesichtern, die visuelle Fehlinformationen erzeugen können. Wir schlagendas erste Abwehrsystem UnGANable gegen GAN-basierte Gesichtsmanipulationen vor, indem wir den Prozess der GAN-Inversion gefährden, der ein wesentlicher Schritt für die anschließende Gesichtsmanipulation ist.  \nAlle Ergebnisse tragen dazu bei, dass die Community einen Einblick in die Datenschutzrisiken von maschinellen Lernmodellen erhält. Wir appellieren an die Gemeinschaft, eine eingehende Untersuchung der Risiken für die Privatsphäre, wie die unsere, im Hinblick auf die sich schnell entwickelnden Techniken des maschinellen Lernens in Betracht zu ziehen.  \nAbstract  \nMachine learning (ML) has made huge progress in the last decade and has been applied to a wide range of critical applications. However, driven by the increasing adoption of machine learning models, the significance of privacy risks has become more crucial than ever. These risks can be classified into two categories depending on the role played by ML models: one in which the models themselves are vulnerable to leaking sensitive information, and the other in which the models are abused to violate privacy.  \nIn this dissertation, we investigate the privacy risks of machine learning models from two perspectives, i.e., the vulnerability of ML models and the abuse of ML models. To study the vulnerability of ML models to privacy risks, we conduct two studies on one of the most severe privacy attacks against ML models, namely the membership inference attack (MIA) . Firstly, we explore membership leakage in lab","cbCainm3ctISa9xX","https://ap.wps.com/l/cbCainm3ctISa9xX","pdf",22312492,1,139,"English","en",105,"# Abstract\n# Background of this Dissertation\n## Research contributions and paper basis","[{\"question\":\"How does the dissertation categorize privacy risks in machine learning models?\",\"answer\":\"It distinguishes two categories: risks arising from model vulnerability that leaks sensitive information, and risks arising from abusing models to violate privacy.\"},{\"question\":\"What is studied first regarding membership inference attacks (MIA)?\",\"answer\":\"The dissertation investigates membership leakage in a label-only scenario, presenting a label-only membership inference attack and showing that leakage is more severe than previously reported.\"},{\"question\":\"What defense is proposed against GAN-based face manipulation and what key step does it target?\",\"answer\":\"It proposes UnGANable, a defense system that jeopardizes GAN inversion, which is an essential step before subsequent face manipulation.\"}]","On the Privacy Risks of Machine Learning Models - Dissertation | PDF",1785726647,350,{"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":86,"head_meta":88,"extra_data":90,"updated_unix":28},"on-the-privacy-risks-of-machine-learning-models-dissertation","",{"@graph":36,"@context":85},[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/on-the-privacy-risks-of-machine-learning-models-dissertation/119849/",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":4},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"How does the dissertation categorize privacy risks in machine learning models?","Question",{"text":75,"@type":76},"It distinguishes two categories: risks arising from model vulnerability that leaks sensitive information, and risks arising from abusing models to violate privacy.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What is studied first regarding membership inference attacks (MIA)?",{"text":80,"@type":76},"The dissertation investigates membership leakage in a label-only scenario, presenting a label-only membership inference attack and showing that leakage is more severe than previously reported.",{"name":82,"@type":73,"acceptedAnswer":83},"What defense is proposed against GAN-based face manipulation and what key step does it target?",{"text":84,"@type":76},"It proposes UnGANable, a defense system that jeopardizes GAN inversion, which is an essential step before subsequent face manipulation.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":53,"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"]