[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-124468-en":3,"doc-seo-124468-105":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},124468,1099514068365,"Aurelia","https://ap-avatar.wpscdn.com/avatar/10000253d8d9f28188e?_k=1776742907772140068",8,"Research & Report","Sense and Sensitivity - Data Utility and User Privacy in Differentially Private Machine Learning","This thesis explores methods for extracting knowledge from data while preserving users’ private information through differentially private machine learning. The core challenge lies in the sensitivity-utility tradeoff when privatizing queries that use vector averages common in gradient-based optimization and data science. The work proposes metric privacy for collaborative model training with distance-dependent guarantees, an online optimization approach for tuning clipping thresholds during training, and efficient tools for auditing privacy in large language model training. It also reframes differential privacy sensitivity to improve utility in federated learning, supported by theory and experiments.","Classe di Scienze  \nCorso di perfezionamento in Data Science  \nXXXVI ciclo  \nSense and Sensitivity:  \nData Utility and User Privacy in Differentially Private Machine Learning  \nSettore Scientifico Disciplinare INF/01  \nCandidato  \nDr. Filippo Galli  \nRelatori  \nProf. Tommaso Cucinotta  \nScuola Superiore Sant’Anna, Pisa, Italy Prof. Catuscia Palamidessi  \nINRIA Saclay; ´Ecole Polytechnique, Paris, France  \nAnno accademico 2023/2024  \nAbstract  \nThis thesis explores existing and novel methods for extracting knowledge from data while preserving users’ private information through differentially private machine learning. The central challenge addressed here is handling the sensitivity-utility tradeoff that arises when privatizing queries involving vector averages, which are found everywhere in gradient-based optimization and data science in general. New approaches are thus proposed to provide researchers and practitioners with additional tools to prioritize the use of one strategy over the other, depending on the specific learning context, the privacy expectations, and the accuracy of the resulting model. First, metric privacy concepts are applied to collaborative model training, providing distance-dependent privacy guarantees without pre-defining sensitivity. An online optimization method is then introduced for tuning the clipping threshold concurrently with model training, reducing privacy exposure and computational requirements while improving utility. Efficient strategies for empirically verifying privacy results in the training of large language models are also developed, encouraging practical privacy auditing. Finally, a new perspective is offered on the definition of differential privacy, suggesting that sensitivity with respect to record replacement rather than addition/removal could yield increased utility in federated learning settings. Through theoretical analyses, algorithms, and experimental evaluations, this work presents ideas and actual techniques for optimizing the privacy-utility tradeoff inherent in differentially private machine learning.  \nList of Figures  \n1.1 Graphical interpretation of the differential privacy definition (inspired by [Meiser, 2018]) ............... 7  \n2.1 Learning federated linear models with: (a, b, c) one ini  \ntial hypothesis and non-sanitized communication,(d, e, f)  \ntwo initial hypotheses and non-sanitized communication,  \n(g, h, i) two initial hypotheses and sanitized communica  \ntion. The first two figures of each row show the parameter  \nvectors released by the clients to the server. The last figure  \nof each row illustrates the trend of the validation loss on clients and data not involved in the optimization...... 37  \n2.2 For the experiment on synthetic data, this figure plots the  \nmax privacy leakage over clients of the same cluster for a  \nround of training. Intervals with constant privacy leakage indicate that the clients with the largest privacy leakage were not sampled (by chance) to participate in those rounds. 38  \n2.3 For the experiment on hospital charge data, this histogram  \nplots the empirical distribution of the privacy budget over  \nthe clients in a particular configuration: ν = 3, 5 initial hypotheses, seed = 3, r is the radius of the neighborhood, and the total number of clients is 2062 ............ 41  \n2.4 RMSE values for models trained with Algorithm 1 on the  \nHospital Charge Dataset. Error bars show the empirical  \nstandard deviation. Lower RMSE values are better for accuracy............................. 42  \n2.5 Effects of the Laplace mechanism in Lemma 2 with different noise multipliers (ref) as a defense strategy against the DLG attack.......................... 44  \n2.6 The first two plots from the left illustrate the spatial dis  \ntribution of the samples in g 1 and g2 , respectively, and the  \nthird plot shows g 1 and g2 superimposed together in the same space.......................... 46  \n2.7 For the experiment with synthetic data, the figure shows  \n","cbCaienSJOXuEijB","https://ap.wps.com/l/cbCaienSJOXuEijB","pdf",5144630,1,129,"English","en",105,"# Abstract\n# Methods for Differential Privacy and Privacy-Utility Tradeoff\n## Metric privacy in collaborative model training\n## Online tuning of clipping thresholds\n## Privacy auditing for large language model training\n## Alternative differential privacy sensitivity definition for federated learning\n# Theoretical Analysis, Algorithms, and Experiments","[{\"question\":\"What is the main problem addressed in this thesis?\",\"answer\":\"The thesis focuses on the sensitivity-utility tradeoff that arises when privatizing queries involving vector averages in differentially private machine learning.\"},{\"question\":\"How does the work improve privacy guarantees in collaborative model training?\",\"answer\":\"It applies metric privacy concepts to collaborative model training, providing distance-dependent privacy guarantees without pre-defining sensitivity.\"},{\"question\":\"What is the contribution of the online optimization method?\",\"answer\":\"An online method tunes the clipping threshold concurrently with model training, reducing privacy exposure and computational requirements while improving utility.\"}]","Sense and Sensitivity - Data Utility and User Privacy in Differentially Private Machine Learning | PDF",1785822559,325,{"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},"sense-and-sensitivity-data-utility-and-user-privacy-in-differentially-private-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/sense-and-sensitivity-data-utility-and-user-privacy-in-differentially-private-machine-learning/124468/",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":4},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What is the main problem addressed in this thesis?","Question",{"text":75,"@type":76},"The thesis focuses on the sensitivity-utility tradeoff that arises when privatizing queries involving vector averages in differentially private machine learning.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does the work improve privacy guarantees in collaborative model training?",{"text":80,"@type":76},"It applies metric privacy concepts to collaborative model training, providing distance-dependent privacy guarantees without pre-defining sensitivity.",{"name":82,"@type":73,"acceptedAnswer":83},"What is the contribution of the online optimization method?",{"text":84,"@type":76},"An online method tunes the clipping threshold concurrently with model training, reducing privacy exposure and computational requirements while improving utility.","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"]