[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-81847-en":3,"doc-seo-81847-105":30,"detail-sidebar-cat-0-en-105":92},{"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":11,"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},81847,5909877438554,"Maeve","https://ap-avatar.wpscdn.com/avatar/5600025385ad2bf12a7?_k=1778553567797529272",8,"Research & Report","Probabilistic Memory for Trustworthy Edge Intelligence","Probabilistic computation underpins trustworthy edge intelligence by quantifying uncertainty, improving robustness, enabling reconstruction, and protecting privacy, yet adoption is constrained by the severe throughput and instruction-overhead gap between Gaussian random number generation and computation. This work proposes probabilistic memory (p-MEM), a unified memory primitive storing distribution parameters and sampling directly at native memory bandwidth, with deterministic data treated as a zero-variance case. A layout-validated p-MEM simulator evaluates device and technology choices, achieving over 1000GSa/s/mm2 GRNG throughput and delivering large reductions in instruction count, latency, and energy for Bayesian neural network workloads.","Probabilistic Memory for Trustworthy Edge Intelligence  \nLikai Pei∗ , Jiahao Zheng∗ , Xueji Zhao∗ , Emilie Ye∗ , Jianbo Liu∗ , Hanqing Tao∗ , Ming-Yen Lee†, Ruiyang Qin‡, Yiyu Shi∗ , Shimeng Yu†, X. Sharon Hu∗ and Ningyuan Cao∗  \nUniversity of Notre Dame  \nNotre Dame, IN, USA  \nGeorgia Institute of Technology  \nAtlanta, GA, USA  \nVillanova University  \nRadnor, PA, USA  \narXiv :2607 .02465v 1 [ cs .AR] 2 Jul 2026  \nABSTRACT  \nProbabilistic computation plays an important role in trustworthy edge intelligence to quantify uncertainty, enhance robustness, reconstruct data and protect privacy, but its adoption is limited by the orders-of-magnitude data throughput gap between Gaussian random number generation (GRNG) and computation, as well as instruction overhead. This paper introduces probabilistic memory (p-MEM), a unified memory primitive that stores distribution parameters (e.g., mean and standard deviation) and samples directly at the native memory bandwidth where deterministic data becomes the zero-variance special case. Using a layout-validated p-MEM simulator, we comprehensively explore device choices, memory specifications, and technology nodes, showing that p-MEM can achieve > 1000GSa/s/mm2 GRNG throughput (including memory arrays access) . Integrated into CPU / GPU systems, p-MEM reduces instruction count by up to 2. 19×/4 . 37×, sampling latency by 562×/3 .45×, and energy by 295. 5×/3 . 53× for Bayesian neural network workloads, providing a scalable hardware substrate for trustworthy probabilistic AI.  \n1 INTRODUCTION  \nAs AI systems are increasingly integrated into real-world edge platforms that directly interact with humans, their trustworthiness– including decision robustness [1], uncertainty awareness [2], partial observation reconstruction [1], and privacy protection [3]–has become a critical requirement. In medical domains such as wearable and implanted devices for ventricular arrhythmia detection [4, 5], blood-glucose monitoring [6], and automated insulin delivery [7], unreliable inference can lead to severe consequences. Similarly, in defense applications such as autonomous drone reconnaissance [8], robustness to sensor noise, out-of-distribution detection, reasoning under uncertainty, and protection of sensitive mission data are equally essential.  \nTo address these challenges, probabilistic computation–a foundational class of algorithms that includes Bayesian neural networks (BNNs) for robust inference [9], variational autoencoders (VAEs) and diffusion models for generative reconstruction [10, 11], probabilistic graphical models and Bayesian decision trees for structured reasoning [12, 13], and differential privacy (DP) for data protection [3, 14–16]-has emerged as a central pillar of trustworthy AI  \nThis paper has been accepted for publication in the proceedings of the ACM/IEEE Design Automation Conference (DAC), 2026 .  \nThe authors are with the College of Engineering at the University of Notre Dame, Notre Dame, IN, USA..  \nFigure 1: (a) Motivation for probabilistic memory to support probabilistic computation in trustworthy edge intelligence. (b) Stateof-the-art GRNG and MAC throughput across platforms and area efficiency. p-MEM is expected to close the gap required for scalable probabilistic AI.  \n(Fig. 1(a)) . These methods explicitly model uncertainty by sampling from learned probability distributions rather than relying on fixed parameters or deterministic operations. However, they require large-scale probabilistic sampling at high sampling intensity–the number of samples required per computation step–which far exceeds traditional uses such as weight initialization. For example, BNNs resample all weights during each inference, and differential privacy injects calibrated noise into every data vector element. This exposes a fundamental hardware bottleneck: existing random number generators (RNGs) are decoupled from the memory arrays that store distribution parameters, incurring significant latency and ener","cbCaig2JmjKlbRT3","https://ap.wps.com/l/cbCaig2JmjKlbRT3","pdf",1657954,6,1,"English","en",105,"# Abstract\n# Introduction\n# Background\n## Probabilistic Computation","[{\"question\":\"What problem does p-MEM target in trustworthy edge intelligence?\",\"answer\":\"The work targets the orders-of-magnitude mismatch between Gaussian random number generation (GRNG) throughput and computation, plus instruction overhead that limits scalable probabilistic AI on edge platforms.\"},{\"question\":\"How does probabilistic memory (p-MEM) differ from existing approaches?\",\"answer\":\"p-MEM stores distribution parameters (such as mean and standard deviation) in memory and performs sampling directly at native memory bandwidth, treating deterministic values as the zero-variance special case.\"},{\"question\":\"What performance benefits does the paper report for p-MEM workloads?\",\"answer\":\"Integrated into CPU/GPU systems, p-MEM reduces instruction count by up to 2.19×/4.37×, sampling latency by 562×/3.45×, and energy by 295.5×/3.53× for representative Bayesian neural network workloads.\"}]","Probabilistic Memory for Trustworthy Edge Intelligence | PDF",1784176625,20,{"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":87,"head_meta":89,"extra_data":91,"updated_unix":28},"probabilistic-memory-for-trustworthy-edge-intelligence","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"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/probabilistic-memory-for-trustworthy-edge-intelligence/81847/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-03","2026-07-16",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"What problem does p-MEM target in trustworthy edge intelligence?","Question",{"text":76,"@type":77},"The work targets the orders-of-magnitude mismatch between Gaussian random number generation (GRNG) throughput and computation, plus instruction overhead that limits scalable probabilistic AI on edge platforms.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does probabilistic memory (p-MEM) differ from existing approaches?",{"text":81,"@type":77},"p-MEM stores distribution parameters (such as mean and standard deviation) in memory and performs sampling directly at native memory bandwidth, treating deterministic values as the zero-variance special case.",{"name":83,"@type":74,"acceptedAnswer":84},"What performance benefits does the paper report for p-MEM workloads?",{"text":85,"@type":77},"Integrated into CPU/GPU systems, p-MEM reduces instruction count by up to 2.19×/4.37×, sampling latency by 562×/3.45×, and energy by 295.5×/3.53× for representative Bayesian neural network workloads.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":93},[94,98,102,106,111,115,120,123,127,130,134],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":46,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},"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":29,"slug":126},9,"Religion & Spirituality","religion-spirituality",{"id":29,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":29,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":46,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":46,"category_name":136,"show_sort_weight":107,"slug":137},19,"General","general"]