[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-121937-en":3,"doc-seo-121937-105":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},121937,137441390410,"Hazel","https://ap-avatar.wpscdn.com/avatar/2000252f4ab5702993?_k=1776741390130283984",8,"Research & Report","On Security and Privacy in Machine Learning","Recent years have seen rapid growth in social media, surveillance cameras, and vision-enabled services, and machine learning usage amplifies privacy and security risks for online users. Privacy laws restrict unfair data use, yet translating legal rules into technical implementations often leaves ambiguity that can enable misuse by data owners and erode end-user privacy. This doctoral work advances privacy-preserving computer vision by extending protections from intrusive facial recognition to object detection and broader vision tasks.","University of Mississippi  \neGrove  \n\n| Electronic Theses and Dissertations | Graduate School |\n| --- | --- |\n| 1-1-2023\u003Cbr>On Security and Privacy in Machine Learning\u003Cbr>Thomas Cilloni\u003Cbr>University of Mississippi\u003Cbr>Follow this and additional works at: [https://egrove.olemiss.edu/etd](https://egrove.olemiss.edu/etd) |  |\n\nRecommended Citation  \nCilloni, Thomas, \"On Security and Privacy in Machine Learning\" (2023) . Electronic Theses and Dissertations. 2746.  \n[https://egrove.olemiss.edu/etd/2746](https://egrove.olemiss.edu/etd/2746)  \nThis Dissertation is brought to you for free and open access by the Graduate School at eGrove. It has been accepted for inclusion in Electronic Theses and Dissertations by an authorized administrator of eGrove. For more information, please [contact egrove@olemiss.edu](contact egrove@olemiss.edu).  \nOn Security and Privacy in Machine  \nLearning  \na Doctoral Dissertation by  \nThomas Cilloni  \nA Doctoral Dissertation submitted in partial  \nfulfillment of the requirements for the degree of  \nDoctor of Philosophy in Engineering Science  \nComputer Science Emphasis  \nin The University of Mississippi  \n2023  \n© Thomas Cilloni 2023 All Rights Reserved  \nABSTRACT  \nIn recent years the use of social media platforms, surveillance cameras, and vision-enabled devices and services has increased exponentially. While the mere use of a device or service is not a privacy issue, Machine Learning use has also drastically increased, and it exposes a number of privacy and security threats, especially towards online users. Privacy legislation around the world prohibits to use of data unfairly, but the conversion of laws to technical applications is often subject to interpretation. Such ambiguity opens a gap for data owners to potentially misuse data, violating end users’ privacy.  \nThe overall objective of this doctoral research is to advance the state-of-the-art in privacypreserving computer vision. Starting from work on privacy protection against intrusive facial recognition models, this research expands first into the realm of object detection, and then further to general Computer Vision tasks. A fast and effective adversarial algorithm is proposed against object detectors. By focusing on sensitive regions of images, this algorithm consistently beats state of the art systems. For general computer vision purposes, this research takes a task-agnostic approach by decoupling the privacy guarantee of data records from the tasks they are used for. This results in a tradeoff between image quality and privacy guarantees, which is measured empirically on example tasks. I discuss three examples of situations where privacy is not currently guaranteed, namely social media posts, surveillance cameras, and self-driving software, and show how each privacy-preserving tool can help mitigate the related residual privacy risk.  \nLIST OF ABBREVIATIONS  \nAGS Automatic Gradient Shielding  \nAML Adversarial Machine Learning  \nASPP Atrous Spatial Pyramid Pooling  \nATM Automated Teller Machine  \nAV Autonomous Vehicle  \nBOV Bag Of Visual words  \nCCPA California Consumer Privacy Act  \nCCPD Chinese City Parking Dataset  \nCCTV Closed-Circuit Television  \nCNN Convolutional Neural Network  \nCOCO [Microsoft] Common Objects in COntext  \nCRF Conditional Random Fields  \nDAG Dense Adversary Generator  \nDDoS Distributed Denial of Service  \nDCNN Deep Convolutional Neural Networks  \nDL Deep Learning  \nDNN Deep Neural Network  \nDSSIM Structural Dissimilarity  \nFaster R-CNN Faster Recurrent Convolutional Neural Network FGSM Fast Gradient Sign Method  \nFHD Full High Definition  \nFPS Frames Per Second  \nGAN Generative Adversarial Network  \nGDPR General Data Protection Regulation  \nHOG Histograms of Oriented Gradients  \nIGS Interactive Gradient Shielding  \nmAP mean Average Precision  \nMI More Imperceptible  \nML Machine Learning  \nNLP Natural Language Processing  \nOCR Optical Character Recognition  \nPascal VOC 2012 Pascal Visual Object Classes 2012  \nPascal VO","cbCailohXc8Eroa6","https://ap.wps.com/l/cbCailohXc8Eroa6","pdf",8348328,1,144,"English","en",105,"# ABSTRACT\n## Motivation and privacy threats\n## Statement of work and scope\n# LIST OF ABBREVIATIONS\n# CONTENT\n## 1. Introduction\n### 1.1 Motivation\n### 1.2 Statement of Work\n## 2. Background\n### 2.1 Introduction\n### 2.2 White-box attacks\n### 2.3 Black-box attacks\n## 3. Face Recognition Privacy","[{\"question\":\"为什么机器学习会带来额外的隐私与安全威胁？\",\"answer\":\"除了设备或服务本身，机器学习的应用会显著扩大隐私和安全风险，尤其影响在线用户。法律约束在落地到技术实现时可能存在理解与解释空间，从而产生漏洞。\"},{\"question\":\"本博士研究的总体目标是什么？\",\"answer\":\"目标是推动隐私保护计算机视觉的最新技术发展。研究从针对侵入式人脸识别模型的隐私保护开始，逐步扩展到目标检测以及更一般的计算机视觉任务。\"},{\"question\":\"该研究提出了哪些隐私保护思路与评估方式？\",\"answer\":\"研究提出针对目标检测器的快速有效对抗算法，聚焦图像敏感区域以在效果上优于现有方法。同时采用任务无关的方式，将数据记录的隐私保证与其使用任务解耦，并在示例任务上以图像质量与隐私保证之间的权衡进行实证测量。\"}]","On Security and Privacy in Machine Learning | PDF",1785807838,363,{"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-security-and-privacy-in-machine-learning","",{"@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-security-and-privacy-in-machine-learning/121937/",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},"为什么机器学习会带来额外的隐私与安全威胁？","Question",{"text":75,"@type":76},"除了设备或服务本身，机器学习的应用会显著扩大隐私和安全风险，尤其影响在线用户。法律约束在落地到技术实现时可能存在理解与解释空间，从而产生漏洞。","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"本博士研究的总体目标是什么？",{"text":80,"@type":76},"目标是推动隐私保护计算机视觉的最新技术发展。研究从针对侵入式人脸识别模型的隐私保护开始，逐步扩展到目标检测以及更一般的计算机视觉任务。",{"name":82,"@type":73,"acceptedAnswer":83},"该研究提出了哪些隐私保护思路与评估方式？",{"text":84,"@type":76},"研究提出针对目标检测器的快速有效对抗算法，聚焦图像敏感区域以在效果上优于现有方法。同时采用任务无关的方式，将数据记录的隐私保证与其使用任务解耦，并在示例任务上以图像质量与隐私保证之间的权衡进行实证测量。","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"]