[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-82740-en":3,"doc-seo-82740-105":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":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},82740,4398048949847,"Eliana","https://ap-avatar.wpscdn.com/avatar/400002536579ef2da7f?_k=1778318612642679267",8,"Research & Report","Scalable Differentially Private Data Compression via Diffusion and Stochastic Codes","Rising volumes of personal data require practical protection of sensitive aspects of identity. Differential privacy (DP) offers formal guarantees, yet compressing high-dimensional outputs such as images under privacy constraints remains difficult because releasing privatized data can be storage intensive and existing methods lack strong compression with guarantees. DP-DiPP combines stochastic codes with diffusion models, enabling direct control of the compression–privacy–utility tradeoff. A Poisson private representation extension encodes privacy-mechanism outputs, and DiffC yields a differentially private image compressor. Experiments on CIFAR-10 show 10–30× better compression with comparable privacy and utility.","Scalable Differentially Private Data Compression via Diffusion and Stochastic Codes  \nGergely Flamich 1 ¨Oyk Sıla Gner 1 Yanxiao Liu 1 Deniz Gndz 1  \narXiv :2607 .03392v 1 [ cs .CR] 3 Jul 2026  \nAbstract  \nThe ever-increasing collection of personal data has created mounting pressure to develop technologies that protect sensitive aspects of individual identity. Differential privacy (DP) provides a principled framework with strong formal guarantees and has already achieved practical success.  \nHowever, releasing high-dimensional data, such as images, has remained elusive: releasing uncompressed privatized data requires significant storage. At the same time, no effective data compression scheme exists that can compress highresolution data with privacy guarantees.  \nWe address this challenge with DP-DiPP, a compression pipeline that combines stochastic codes with diffusion models. DP-DiPP is highly flexible:  \nthe practitioner has direct control over the compression rate-privacy-utility tradeoff. As the theoretical backbone, we extend the Poisson private representation (PPR) of (Liu et al., 2024) to encode the outputs of privacy mechanisms. We then combine it with DiffC, a diffusion-based lossy data compression method, to obtain a differentially private image compressor. Our experiments on privatized image classification on CIFAR-10 demonstrate that DP-DiPP significantly outperforms the baseline, achieving a 10-30 times better compression while retaining comparable privacy guarantees and utility.  \n1. Introduction  \nOver the past decades, the rapid growth in data collection and its use in downstream systems such as machine learning has produced undeniable benefits. Yet, this has also raised serious concerns about the identifiability of sensitive information and individual privacy. Therefore, developing frameworks and algorithms that protect privacy is essential  \n1Imperial College London, London, UK. Correspondence to: Gergely Flamich \u003C[g.flamich@imperial.ac.uk](g.flamich@imperial.ac.uk)>.  \nPreprint. July 7, 2026.  \nto ensure user trust. In this paper, we adopt differential privacy (DP) as the privacy framework, a stronger variant of DP (DP; Dwork et al., 2006 ; Dwork & Roth, 2014) . DP has already found practical success, such as for collecting user data (Erlingsson et al., 2014), for releasing privatized tabular data, like census data (United States Census Bureau, 2021), and for preventing generative models from memorizing and revealing their training sets (Liu et al., 2023) .  \nLocal DP (LDP) protects individuals’ data by requiring that a system’s output reveal only limited information about its input. In practice, this idea translates to randomizing the sensitive data X before revealing it to an untrusted party, according to a privacy mechanism. For example, if X represents location data, such as GPS coordinates (Andrs et al., 2013), one possible privacy mechanism is to add appropriately calibrated noise η to the coordinates to prevent precise localization, that is, Y = X + η . Of course, we still wish to use the privatized data for some purpose. Hence, a natural tradeoff arises between the privatized data’s utility for its downstream task and its privacy guarantee.  \nHowever, despite its success for tabular data, scaling DP to high-dimensional data such as images, audio, or video remains challenging. This difficulty stems from several factors. First, directly releasing privatized high-dimensional data can incur substantial storage and communication costs. Second, the noise required to provide meaningful privacy guarantees can severely degrade the data’s utility in highdimensional domains. To the best of our knowledge, prior work has not systematically studied high-resolution privatized data compression, where the goal is to jointly preserve privacy, reduce storage cost, and retain downstream utility. This is the challenge we tackle in this paper.  \nThe theoretical foundation of our work is in the emerging field of data compr","cbCaitiQxiPqR2hg","https://ap.wps.com/l/cbCaitiQxiPqR2hg","pdf",425481,2,1,13,"English","en",105,"# Abstract\n# Introduction\n## Differential privacy and local DP\n## Challenges of compressing high-dimensional privatized data\n## Stochastic codes for randomized lossy compression\n## Poisson private representation and its limitations\n## DP-DiPP approach combining extended PPR with diffusion compression","[{\"question\":\"What problem does DP-DiPP address?\",\"answer\":\"DP-DiPP targets the challenge of compressing high-dimensional privatized data, such as images, while preserving differential privacy guarantees and retaining downstream utility.\"},{\"question\":\"How does DP-DiPP combine stochastic codes and diffusion models?\",\"answer\":\"DP-DiPP builds a compression pipeline that extends Poisson private representation to encode privacy-mechanism outputs, then integrates it with DiffC, a diffusion-based lossy compression method, to produce a differentially private image compressor.\"},{\"question\":\"What results are reported for the CIFAR-10 privatized image classification experiments?\",\"answer\":\"The experiments show DP-DiPP significantly improves compression by a factor of about 10–30 while maintaining comparable privacy guarantees and utility relative to the baseline.\"}]",1784182606,33,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"scalable-differentially-private-data-compression-via-diffusion-and-stochastic-codes","",{"@graph":36,"@context":85},[37,53,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,47,50],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":20},"https://docshare.wps.com/document/","Document",{"item":48,"name":12,"@type":43,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":43,"position":52},"https://docshare.wps.com/document/scalable-differentially-private-data-compression-via-diffusion-and-stochastic-codes/82740/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":24,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":41,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-23","2026-07-16",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 problem does DP-DiPP address?","Question",{"text":75,"@type":76},"DP-DiPP targets the challenge of compressing high-dimensional privatized data, such as images, while preserving differential privacy guarantees and retaining downstream utility.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does DP-DiPP combine stochastic codes and diffusion models?",{"text":80,"@type":76},"DP-DiPP builds a compression pipeline that extends Poisson private representation to encode privacy-mechanism outputs, then integrates it with DiffC, a diffusion-based lossy compression method, to produce a differentially private image compressor.",{"name":82,"@type":73,"acceptedAnswer":83},"What results are reported for the CIFAR-10 privatized image classification experiments?",{"text":84,"@type":76},"The experiments show DP-DiPP significantly improves compression by a factor of about 10–30 while maintaining comparable privacy guarantees and utility relative to the baseline.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"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"]