[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-121456-en":3,"doc-seo-121456-105":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},121456,13056703019662,"Evangeline","https://ap-avatar.wpscdn.com/avatar/be000253a8e92610077?_k=1778726343310543188",8,"Research & Report","Towards Trustworthy Machine Learning Models in Vision, Physics, and Language Applications - Thesis","Machine learning, especially deep neural networks, is increasingly deployed in safety-critical domains such as autonomous driving, medical diagnosis, finance, and manufacturing. Although driven by strong benchmark performance, these data-driven models can behave unpredictably under non-standard inputs, creating risks when humans interact with systems or when failures occur. This thesis develops methods for trustworthy machine learning by improving probabilistic certification, enabling partial-derivative verification for neural networks, and assessing safety risks in fine-tuning large language models. It also examines how open-source generative AI can support trustworthiness through broader implications.","Towards Trustworthy Machine Learning Models in Vision, Physics, and Language Applications  \nFrancisco Girbal Eiras  \nLinacre College  \nUniversity of Oxford  \nA thesis submitted for the degree of Doctor of Philosophy  \nTrinity Term 2024  \nAbstract  \nThe wide-ranging impact of machine learning, particularly deep neural networks, cannot be overstated. These highly capable models are now deployed in critical domains such as autonomous driving, medical diagnosis, finance, and manufacturing. While their adoption is driven by superior performance on benchmark tasks, their data-driven nature often renders them unpredictable when encountering non-standard inputs. This unpredictability poses a significant challenge in safety-critical applications, where interactions with humans or the potential for system failures could lead to severe consequences. This underscores the need for trustworthy machine learning, where models must not only excel in standard metrics but also prove to be reliable and robust in real-world settings.  \nOur work addresses this need by improving methods that aim to either certify the robustness of these systems or, at a minimum, provide strong empirical evaluations of robustness and safety to support responsible deployment. We present advancements in probabilistic certification for image classification via randomized smoothing, introduce a general framework for verifying the partial derivatives of neural networks, which has applications in certifying the correctness of physics-informed neural networks, and analyse the safety risks involved in fine-tuning large language models on task-specific data, along with mitigation strategies. Additionally, we explore the broader implications of open-source generative AI models for improving trustworthiness. These contributions mark a step forward in developing trustworthy machine learning systems, and we conclude by discussing their strengths, limitations, as well as key open questions that remain for the field.  \nTowards Trustworthy Machine Learning Models in Vision, Physics, and Language  \nApplications  \nFrancisco Girbal Eiras  \nLinacre College University of Oxford  \nA thesis submitted for the degree of Doctor of Philosophy  \nTrinity Term 2024  \nAcknowledgements  \nIn the final year of my DPhil, I found myself drawn to the works of Kurt Vonnegut, particularly his concept of the karass in Cat’s Cradle—–a group of people cosmically bound together, often unknowingly, to accomplish a shared purpose. Reflecting on this idea, I realize how many people have been part of my own karass throughout this journey. While they may not have known it at the time, each has played a crucial role in helping me reach this point, and for that, I am deeply grateful.  \nMy DPhil journey would have been impossible without the guidance and encouragement of my supervisors and mentors. First, I am deeply thankful to M. Pawan Kumar, who believed in me from the beginning and helped me get started with my PhD. Even after transitioning to DeepMind, Pawan continued to provide guidance and insightful feedback across all my projects, always offering invaluable advice. He taught me perhaps the simplest yet most important lesson that I will take away from my DPhil: to do impactful research, start from an important problem and work towards a methodology that solves it, not the other way around. I am also incredibly grateful to Philip Torr, who, despite his busy schedule, always made time to share his wisdom and offer mentoring. Our lunches at Catz will remain a memorable source of inspiration. I must also thank my co-supervisor, Adel Bibi, whose consistent presence throughout my projects was a cornerstone of my progress. His feedback, endless discussions, and readiness to help whenever I hit a wall were essential to my growth. Though not officially my supervisor, Puneet K. Dokania’s mentorship and guidance during my internships at FiveAI was indispensable, and I greatly appreciate his support. I would also like to","cbCairBWIWPS1J6t","https://ap.wps.com/l/cbCairBWIWPS1J6t","pdf",7729186,1,239,"English","en",105,"# Abstract\n# Acknowledgements\n## Supervisors and mentors\n## Financial support and collaborators\n## Lab colleagues and friends","[{\"question\":\"Why is trustworthy machine learning needed in safety-critical applications?\",\"answer\":\"Machine learning models can achieve high benchmark accuracy but remain unpredictable when encountering non-standard inputs, which can lead to severe consequences in safety-critical settings involving human interaction or potential system failures.\"},{\"question\":\"What contributions does the thesis make for improving model trustworthiness?\",\"answer\":\"It advances probabilistic certification for image classification using randomized smoothing, introduces a framework to verify neural network partial derivatives, and analyzes safety risks and mitigation strategies for fine-tuning large language models on task-specific data.\"},{\"question\":\"How does the thesis connect trustworthiness with open-source generative AI?\",\"answer\":\"It explores broader implications of open-source generative AI models as a pathway to improving trustworthiness, alongside the presented certification and safety-evaluation methods.\"}]","Towards Trustworthy Machine Learning Models in Vision, Physics, and Language Applications - Thesis | PDF",1785735747,602,{"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-models-in-vision-physics-and-language-applications-thesis","",{"@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-models-in-vision-physics-and-language-applications-thesis/121456/",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},"Why is trustworthy machine learning needed in safety-critical applications?","Question",{"text":75,"@type":76},"Machine learning models can achieve high benchmark accuracy but remain unpredictable when encountering non-standard inputs, which can lead to severe consequences in safety-critical settings involving human interaction or potential system failures.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"What contributions does the thesis make for improving model trustworthiness?",{"text":80,"@type":76},"It advances probabilistic certification for image classification using randomized smoothing, introduces a framework to verify neural network partial derivatives, and analyzes safety risks and mitigation strategies for fine-tuning large language models on task-specific data.",{"name":82,"@type":73,"acceptedAnswer":83},"How does the thesis connect trustworthiness with open-source generative AI?",{"text":84,"@type":76},"It explores broader implications of open-source generative AI models as a pathway to improving trustworthiness, alongside the presented certification and safety-evaluation methods.","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"]