[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-119043-en":3,"doc-seo-119043-105":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},119043,137441390410,"Hazel","https://ap-avatar.wpscdn.com/avatar/2000252f4ab5702993?_k=1776741390130283984",8,"Research & Report","RQP-SGD - Differential Private Machine Learning through Noisy SGD and Randomized Quantization","IoT expansion drives the need for edge machine learning with real-time, efficient, and secure processing. When models require quantized discrete weights, privacy of the underlying dataset must still be preserved. RQP-SGD provides privacy-preserving quantization by combining differential privacy stochastic gradient descent (DP-SGD) with randomized quantization, yielding measurable privacy guarantees. The work analyzes utility convergence for convex objectives under quantization constraints and validates performance advantages over deterministic quantization through experiments on two datasets.","arXiv :2402 .06606v 1 [ cs .LG] 9 Feb 2024  \nRQP-SGD: DIFFERENTIAL PRIVATE MACHINE LEARNING THROUGH NOISY SGD AND RANDOMIZED QUANTIZATION ∗  \nCe Feng, Parv Venkitasubramaniam  \nDepartment of Electrical and Computer Engineering  \nLehigh University  \nCity  \n{cef419, [pav309}@lehigh.edu](pav309}@lehigh.edu)  \nABSTRACT  \nThe rise of IoT devices has prompted the demand for deploying machine learning at-the-edge with real-time, efficient, and secure data processing. In this context, implementing machine learning (ML) models with real-valued weight parameters can prove to be impractical particularly for large models, and there is a need to train models with quantized discrete weights. At the same time, these low-dimensional models also need to preserve privacy of the underlying dataset. In this work, we present RQP-SGD, a new approach for privacy-preserving quantization to train machine learning models for low-memory ML-at-the-edge. This approach combines differentially private stochastic gradient descent (DP-SGD) with randomized quantization, providing a measurable privacy guarantee in machine learning. In particular, we study the utility convergence of implementing RQP-SGDon ML tasks with convex objectives and quantization constraints and demonstrate its efficacy over deterministic quantization. Through experiments conducted on two datasets, we show the practical effectiveness of RQP-SGD.  \n1 Introduction  \nAs IoT devices proliferate across industries, there is an increasing demand to process data closer to the source [1, 2], enabling real-time insights and decision-making. There is now a growing need for machine learning (ML) at-the-edge, where ML models are deployed directly onto IoT devices or gateways, enabling them to perform data analysis and make predictions locally, without the need to transmit raw data to centralized servers. This approach not only reduces latency and conserves network resources but can also enhance privacy and security.  \nAlthough ML algorithms have shown tremendous success in various domains, ML at the edge brings unique constraints in the limited dimensionality of the models learned, and the need for strong privacy guarantees, particularly for IoT deployed in sensitive applications such as health monitoring and energy management. The objective of this work is to propose a joint privacy-preserving quantization approach to train neural networks for ML-at-the-edge IoT applications.  \nPast studies [3–6] propose different approaches to guaranteeing privacy in ML, notably through the concept of differential privacy [7] -a widely accepted quantitative measure of privacy. Specifically, these methods rely on a noisy training approach known as Differentially Private Stochastic Gradient Descent (DP-SGD). DP-SGD [8] directly perturbs the gradient at each descent update with random noise drawn from the Gaussian distribution, resulting in a significant impact on utility. Several recent papers propose noise reduction methods for the differentially private noise added to the reduced gradient. For instance, recent studies by [4, 9, 10] have explored methods for applying DP-SGD to gradients in a reduced dimensional space. Another work [11] performs differentially private perturbations in the spectral domain and introduces filtering for noise reduction.  \nIn this work, we propose Randomized Quantization Projection-Stochastic Gradient Descent (RQP-SGD), a new approach to achieving differential privacy in ML when weights need to be discretized. Our work is motivated by the knowledge that quantization is a process of removing redundant information by converting model parameters from  \n∗This work is accepted by the 5th AAAI Workshop on Privacy-Preserving Artificial Intelligence.  \nhigh-precision to low-precision representations. Since privacy also requires the removal of sensitive information, our approach exploits the synergies between these two processes. Furthermore, quantization effectively reduces the memory and computat","cbCaibWJtpbydoZ5","https://ap.wps.com/l/cbCaibWJtpbydoZ5","pdf",465101,1,11,"English","en",105,"# Introduction\n## Background: Edge ML and Privacy Needs\n## Differential Privacy and DP-SGD Noise\n## Proposed Method: Randomized Quantization Proj-SGD\n## Main Contributions\n## Experiments and Utility-Privacy Trade-off","[{\"question\":\"What problem does RQP-SGD address in IoT edge machine learning?\",\"answer\":\"It addresses training ML models at the edge using quantized discrete weights while still preserving differential privacy of the underlying dataset.\"},{\"question\":\"How does RQP-SGD combine privacy and quantization?\",\"answer\":\"RQP-SGD combines DP-SGD, which adds Gaussian noise to gradients, with randomized quantization so that the discretization process supports privacy while reducing memory and computation.\"},{\"question\":\"What evidence shows RQP-SGD can outperform deterministic quantization?\",\"answer\":\"The paper analyzes the utility-privacy trade-off theoretically and demonstrates improved results in experiments on MNIST and the Breast Cancer Wisconsin (Diagnostic) dataset, including a reported 35% higher accuracy on the Diagnostic dataset while maintaining (1.0, 0)-DP.\"}]","RQP-SGD - Differential Private Machine Learning through Noisy SGD and Randomized Quantization | PDF",1785722062,28,{"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},"rqp-sgd-differential-private-machine-learning-through-noisy-sgd-and-randomized-quantization","",{"@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/rqp-sgd-differential-private-machine-learning-through-noisy-sgd-and-randomized-quantization/119043/",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},"What problem does RQP-SGD address in IoT edge machine learning?","Question",{"text":75,"@type":76},"It addresses training ML models at the edge using quantized discrete weights while still preserving differential privacy of the underlying dataset.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does RQP-SGD combine privacy and quantization?",{"text":80,"@type":76},"RQP-SGD combines DP-SGD, which adds Gaussian noise to gradients, with randomized quantization so that the discretization process supports privacy while reducing memory and computation.",{"name":82,"@type":73,"acceptedAnswer":83},"What evidence shows RQP-SGD can outperform deterministic quantization?",{"text":84,"@type":76},"The paper analyzes the utility-privacy trade-off theoretically and demonstrates improved results in experiments on MNIST and the Breast Cancer Wisconsin (Diagnostic) dataset, including a reported 35% higher accuracy on the Diagnostic dataset while maintaining (1.0, 0)-DP.","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"]