[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-122359-en":3,"doc-seo-122359-105":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":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},122359,1374391974468,"Eden","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"Research & Report","Towards Trustworthy Machine Learning - Dissertation","In an era dominated by ubiquitous machine learning, trust becomes central for individuals, organizations, and society. This dissertation targets trustworthiness in ML beyond raw performance, emphasizing privacy, fairness, and robustness. It demonstrates prompt-based data extraction from large language models and proposes an efficient defense without model re-training. It also introduces fairness training with limited demographic information, and develops lightweight defenses for federated learning against backdoor attacks, analyzing how local distribution shifts can degrade fairness and robustness faster than accuracy.","TOWARDS TRUSTWORTHY MACHINE LEARNING  \nby  \nMustafa Safa Ozdayi  \nAPPROVED BY SUPERVISORY COMMITTEE:  \n\n| Murat Kantarcioglu, Chair |\n| --- |\n| Yulia Gel |\n| Rishabh Iyer |\n\nLatifur Khan  \nCopyright © 2023 Mustafa Safa Ozdayi All rights reserved  \nDedicated to annem, babam, halam, and kardeslerim.  \nTOWARDS TRUSTWORTHY MACHINE LEARNING  \nby  \nMUSTAFA SAFA OZDAYI, BS, MS  \nDISSERTATION Presented to the Faculty of The University of Texas at Dallas in Partial Fulfillment  \nof the Requirements for the Degree of  \nDOCTOR OF PHILOSOPHY IN  \nCOMPUTER SCIENCE  \nTHE UNIVERSITY OF TEXAS AT DALLAS December 2023  \nACKNOWLEDGMENTS  \nDoing a PhD and writing a dissertation is not the easiest thing in the world. Yet, having a good advisor helps. First and foremost, I would like to thank my advisor, Murat Kantarcioglu, for having guided me throughout this journey and allowing me to explore my research interests freely. I am indebted to his patience.  \nApart from having a good advisor, having good collaborators is a blessing as well. I would like to thank Harsh Desai, Shihabul Islam, Yue Guo, Yulia Gel, Rishabh Iyer, and Bradley Malin for our productive collaborations.  \nI also consider myself very lucky to have done a few industry internships during my PhD. In particular, I would like to thank Mahdi Zamani of Visa Research, Swanand Kadhe of IBM Research, and Charith Peris of Amazon for being great mentors and collaborators during my time at industry.  \nObviously, all work and no play makes a PhD student a dull boy. I am happy to have madea few good friends during my studies to blow off some steam. In no particular order, I’d like to thank my friends Ceren, Cuneyt, Chengen, Firat, Vibha, Omer Faruk Sr., Omer Faruk Jr. , Aref, Ali, and Imrul.  \nFinally, I cannot truly express my gratitude for my family. Annem, obviously you played the most vital role in my academic life, and elsewhere. You instilled a curiosity for science which has led me all the way to a PhD. I literally would not be here without you. Babam, thanks for letting me know that I can always count on you. Halam, I am forever grateful foryour support. Kardeslerim, you guys are fine too! Really!  \nOctober 2023  \nTOWARDS TRUSTWORTHY MACHINE LEARNING  \nMustafa Safa Ozdayi, PhD  \nThe University of Texas at Dallas, 2023  \nSupervising Professor: Murat Kantarcioglu, Chair  \nIn an era marked by ubiquitous machine learning (ML) applications, the question of trust has risen to the forefront of concern. From financial institutions using ML for credit risk modeling to the rapid adoption of Large Language Models (LLMs) such as ChatGPT, the reliance on ML systems has become a defining characteristic of our lives. However, as ML’s influence grows, so do the implications of trust; affecting individuals, organizations, and society at large. Trustworthiness in ML transcends mere performance; it involves the intricate balance of privacy, fairness, robustness, and more. This dissertation addresses these core issues, aiming to enhance the trustworthiness and reliability of ML applications.  \nPrivacy is a paramount concern in the context of LLMs. Concretely, LLMs possess the ability to memorize segments of their training data, and can reproduce memorized content when given appropriate prompts. This becomes particularly significant when models are trained on data containing sensitive and private information. In our work, we show how we can discover prompts capable of eliciting memorized content from LLMs which corresponds to a data extraction attack. Additionally, by deriving valuable insights from our attack, we create a defense mechanism that reduces the chances of an LLM generating memorized content. Our defense is efficient as it does not need re-training of models, and offers adjustable privacy-utility trade-offs.  \nFairness in ML is another critical area as models increasingly inform decision-making processes. Current research highlights that, models may amplify and propagate societal biases encoded ","cbCaicuAaq5laFJK","https://ap.wps.com/l/cbCaicuAaq5laFJK","pdf",8514954,1,114,"English","en",105,"# ACKNOWLEDGMENTS\n# ABSTRACT\n# LIST OF FIGURES\n# LIST OF TABLES\n# CHAPTER 1 INTRODUCTION\n# CHAPTER 2 BACKGROUND","[{\"question\":\"What main trust issues does the dissertation focus on?\",\"answer\":\"It focuses on privacy, fairness, and robustness in machine learning systems, especially as ML systems increasingly influence high-impact decisions and deployments.\"},{\"question\":\"How does the work address privacy risks in large language models?\",\"answer\":\"It shows how prompts can elicit memorized training content from LLMs, and then derives a defense mechanism that reduces memorization leakage while allowing adjustable privacy-utility trade-offs.\"},{\"question\":\"What is proposed to improve federated learning security?\",\"answer\":\"The dissertation proposes a lightweight defense against backdoor attacks in federated learning, and reports empirical evidence of improved effectiveness across multiple settings.\"}]","Towards Trustworthy Machine Learning - Dissertation | PDF",1785810220,287,{"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},"towards-trustworthy-machine-learning-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/towards-trustworthy-machine-learning-dissertation/122359/",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","2026-08-04",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 trust issues does the dissertation focus on?","Question",{"text":75,"@type":76},"It focuses on privacy, fairness, and robustness in machine learning systems, especially as ML systems increasingly influence high-impact decisions and deployments.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the work address privacy risks in large language models?",{"text":80,"@type":76},"It shows how prompts can elicit memorized training content from LLMs, and then derives a defense mechanism that reduces memorization leakage while allowing adjustable privacy-utility trade-offs.",{"name":82,"@type":73,"acceptedAnswer":83},"What is proposed to improve federated learning security?",{"text":84,"@type":76},"The dissertation proposes a lightweight defense against backdoor attacks in federated learning, and reports empirical evidence of improved effectiveness across multiple settings.","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"]