[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-290546-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-290546-en":131},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","linear-transformations-for-privacy-preserving-machine-learning","Linear Transformations for Privacy-Preserving Machine Learning","","Proposal and evaluation of two privacy-preserving encoding methods for machine learning: Random Walk Feature Encoder and Fast Privacy Encoder. Random Walk performs sequential linear transformations driven by a secret walk path over features and stochastic weights sampled from Beta distributions, preserving linear relationships while making reconstruction attacks computationally infeasible because the transformation matrix cannot be learned without the secret path. Fast Privacy uses hash-based feature expansion with polynomial transformations and noise injection to balance privacy and utility. Includes theoretical attack-complexity analysis and empirical evaluation against realistic linear reconstruction attacks, positioning the approach as an alternative to differential privacy and homomorphic encryption for efficient privacy protection.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/linear-transformations-for-privacy-preserving-machine-learning/290546/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/linear-transformations-for-privacy-preserving-machine-learning/290546.png","ImageObject",300,407,{"name":42,"@type":43},"Noah","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-11-08","2026-09-17",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",14,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What are the two privacy-preserving encoding methods proposed in the paper?","Question",{"text":63,"@type":64},"The paper proposes Random Walk Feature Encoder and Fast Privacy Encoder for machine learning data encoding.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How does the Random Walk method make reconstruction attacks computationally infeasible?",{"text":68,"@type":64},"It applies sequential linear transformations using a secret walk path and stochastic weights sampled from Beta distributions, so the transformation matrix cannot be learned without knowing the secret path and weights.",{"name":70,"@type":61,"acceptedAnswer":71},"In what way is Fast Privacy designed to balance privacy and utility?",{"text":72,"@type":64},"It uses hash-based feature expansion with polynomial transformations plus noise injection, aiming to provide a practical privacy-utility trade-off.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},290546,1789642001,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,115,120,123,127],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":112,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":113,"slug":114},8,30,"research-report",{"id":116,"doc_module":4,"doc_module_name":25,"category_name":117,"show_sort_weight":118,"slug":119},9,"Religion & Spirituality",20,"religion-spirituality",{"id":118,"doc_module":4,"doc_module_name":25,"category_name":121,"show_sort_weight":118,"slug":122},"World Cup","world-cup",{"id":124,"doc_module":4,"doc_module_name":25,"category_name":125,"show_sort_weight":124,"slug":126},10,"Lifestyle","lifestyle",{"id":128,"doc_module":4,"doc_module_name":25,"category_name":129,"show_sort_weight":97,"slug":130},19,"General","general",{"code":4,"msg":82,"data":132},{"doc_id":79,"user_id":133,"nickname":42,"user_avatar":134,"doc_module":4,"category_id":112,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":33,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":80,"read_time":124,"preview_page_count":4},8796095462418,"https://ap-avatar.wpscdn.com/avatar/80000253c1241d02b47?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778826106357471780","© 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.  \nLinear Transformations for Privacy-Preserving  \nMachine Learning  \nPavel Sulimov  \nInstitute of Computer Science Zurich University of Applied Sciences Winterthur, Switzerland [suli@zhaw.ch](suli@zhaw.ch)  \nSergey Eremeykin  \nBanco Plata Limassol, Cyprus [eremeykins@gmail.com](eremeykins@gmail.com)  \nAbstract—We propose and evaluate two privacy-preserving encoding methods for machine learning: Random Walk Feature Encoder and Fast Privacy Encoder. The Random Walk method is a novel approach that applies sequential linear transformations through a secret walk path over features, creating complex dependencies by visiting features in a secret order and applying transformations with secret weights sampled from Beta distributions. This sequential application preserves linear relationships needed for machine learning while making reconstruction attacks computationally infeasible, as the transformation matrix cannot be learned without knowledge of the secret walk path and weights. The Fast Privacy method uses hash-based feature expansion with polynomial transformations and noise injection, providing a practical alternative that balances privacy and utility. We provide a theoretical analysis of attack complexity for the Random Walk method, demonstrating exponential brute-force complexity, and evaluate both methods against realistic linear reconstruction attacks. Our approaches offer alternatives to differential privacy and homomorphic encryption, balancing privacy protection with computational efficiency and model utility. The Random Walk method is particularly well-suited for neural networks with linear first layers, as it preserves the linear structure that these models rely on while providing meaningful privacy guarantees.  \nIndex Terms—privacy-preserving machine learning, data obfuscation, random walk, linear transformations, reconstruction attacks  \nI. INTRODUCTION  \nMachine learning (ML) models are increasingly deployed on data containing sensitive personal information. Protecting this data during training and inference is a central requirement in many applications. Traditional privacy-preserving techniques often fall into two extremes: they either provide strong theoretical guarantees at the cost of significant utility (e.g., differential privacy), or offer high security with prohibitive computational overhead (e.g., homomorphic encryption) .  \nA core challenge in ethical machine learning is that sensitive attributes (static like birth date, or dynamic like interaction history) are often essential for accuracy in fields from healthcare to recommendations, yet their direct use creates privacy and fairness risks, leading to legal and ethical restrictions. Weresolve this by enabling data to be transformed into an encoded representation at the point of collection (e.g., on the user’s device) . The raw data is never stored, but the encoded version preserves predictive utility for downstream models. This allows organizations to build accurate, personalized systems  \nwhile technically ensuring compliance with privacy and nondiscrimination principles by design.  \nTabular demographic data, comprising features such as age, income, and occupation, is particularly exposed to reconstruction and identification attacks. Unlike image or text data, where local structure is often redundant, tabular features can carry high individual information content. In this paper, we propose a middle-ground approach using sequential linear transformations to obfuscate individual feature values while preserving the global structure required for accurate machine learning predictions. ","cbCaifkMS5uL2KTP","https://ap.wps.com/l/cbCaifkMS5uL2KTP","pdf",662830,"English","# Abstract\n# I. Introduction\n# II. Related Work\n# III. Proposed Methods\n## A. Random Walk Feature Encoder","[{\"question\":\"What are the two privacy-preserving encoding methods proposed in the paper?\",\"answer\":\"The paper proposes Random Walk Feature Encoder and Fast Privacy Encoder for machine learning data encoding.\"},{\"question\":\"How does the Random Walk method make reconstruction attacks computationally infeasible?\",\"answer\":\"It applies sequential linear transformations using a secret walk path and stochastic weights sampled from Beta distributions, so the transformation matrix cannot be learned without knowing the secret path and weights.\"},{\"question\":\"In what way is Fast Privacy designed to balance privacy and utility?\",\"answer\":\"It uses hash-based feature expansion with polynomial transformations plus noise injection, aiming to provide a practical privacy-utility trade-off.\"}]","Linear Transformations for Privacy-Preserving Machine Learning | PDF"]