[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-127473-en":3,"doc-seo-127473-105":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":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},127473,962084925290,"Ophelia","https://ap-avatar.wpscdn.com/davatar_085a072bc5b1113ac321206ff7593b45",8,"Research & Report","CostSphere - A Cost Model-Driven Privacy-Preserving Machine Learning Framework with Network Context Adaptation","CostSphere is a cost-model-driven privacy-preserving machine learning (PPML) framework addressing inefficiencies of fixed hybrid protocol assignments across varying network environments. It dynamically assigns Fully Homomorphic Encryption (FHE) and Secure Multi-Party Computation (MPC) protocols to optimize inference efficiency under changing LAN/WAN conditions. Using a predictive cost model built on MLIR’s TOSA-level dialect and an ILP-based solver, CostSphere achieves robust performance for Transformer models, delivering 6.68×–12.92× inference-runtime improvements over state-of-the-art baselines.","CostSphere: A Cost Model-Driven Privacy-Preserving Machine Learning Framework with Network Context Adaptation  \nYuntian Chen, Zhanyong Tang, Tianpei Lu, Bingsheng Zhang, Zhiying Shi, Zhiyuan Ning  \nPrivacy-preserving machine learning (PPML) is critical for protecting sensitive data in domains like healthcare, finance, and recommendation systems. Fully Homomorphic Encryption (FHE) and Secure Multi-Party Computation (MPC) are key enablers of secure computation, yet existing hybrid approaches often suffer from fixed protocol assignments, resulting in inefficiencies across diverse network environments, such as LANsand WANs. To address this, we introduce COSTSPHERE, a cost-model-driven framework that dynamically assigns FHE and MPC protocols to optimize computational efficiency under varying network conditions. Utilizing a predictive cost model based on MLIR’s TOSA-level dialect and an ILP-based solver, COSTSPHERE ensures robust performance for Transformerbased models. Experimental results demonstrate that COSTSPHERE delivers 6.68 × to 12.92 × improvements in inference runtime compared to state-of-the-art solutions like BumbleBee (NDSS ’25), enabling scalable and network-agnostic PPML across diverse computational scenarios.  \nI. INTRODUCTION  \nThe widespread adoption of machine learning across domains such as healthcare, finance, and recommendation systems [1]–[3] has intensified concerns about data privacy, particularly when handling sensitive personal information [4] . To address these concerns, Privacy-Preserving Machine Learning (PPML) [5], [6], [7], [8], [9] has emerged as a critical field, leveraging techniques like Secure Multi-Party Computation (MPC) [10] and Fully Homomorphic Encryption (FHE) [11],[12] . MPC enables n participants, each with private inputs x 1 ,..., xn , to compute a function f(x1 ,..., xn) while revealing only the output. FHE allows computations on encrypted data, producing encrypted results that can be decrypted only by the private key holder, ensuring equivalence to plaintext computations without data leakage.  \nIn practical PPML applications, particularly for complex models like Transformers [13], computations involve both large-scale linear operations (e.g., matrix multiplications) and non-linear operations (e.g., SoftMax, Tanh, ReLU) [14] . Efficiently managing these diverse operations requires the synergistic use of MPC and FHE to exploit their complementary strengths. Recent studies have explored manually optimized hybrid protocols to enhance computational efficiency in PPML, such as ABY [15], ABY3 [16], CostCo [17], and [18]–[21], addressing the challenges posed by complex computational loads and non-linear operations. However, these  \nCryptFlow2 vs Nexus: LAN/WAN  \nRuntime  \n30  \n25  \n20  \n15  \n10  \n5  \n0  \n\n|  |  CryptFlow2 (Linear) |  25.68 |  |\n| --- | --- | --- | --- |\n|  |  |  |  |\n|  Nexus (Linear) |  |  |  |\n|  |  |  |  |\n| 13.10  |  |  | 13.10 |\n|  |  |  |  |\n| 2.56  |  |  |  |\n\nLAN WAN  \n(a) Single technique for Matrix  \nRuntime  \n30  \n25  \n20  \n15  \n10  \n5  \n0  \nBOLT vs Nexus: LAN/WAN  \n\n|  |  BOLT (Softmax) |  |  |  |\n| --- | --- | --- | --- | --- |\n|  |  |  |  |  23.00 |\n|  |  |  |  |  |\n|  |  |  |  |  |\n| 5.51  5.51 |  |  |  |  |\n| 1.96  |  |  |  |  |\n\nLAN WAN  \n(b) Single technique for Softmax  \niugltiiatoionmp. arison of the perfunctiormanc. of a single technology in different deployment environments. WAN={100mbps bandwidth, 80ms latency} . The green markings indicate the use of MPC technology, while the red ones indicate the use of FHE technology.  \napproaches often lack adaptability to varying network conditions, limiting their effectiveness in dynamic environments.  \nCurrent research on privacy-preserving inference for Transformer-based models, such as BERT [22] and GPT [23], often employs a fragmented approach to assigning Fully Homomorphic Encryption (FHE) and Multi-Party Computation (MPC) protocols. Works like BOLT [24] and BumbleBee [25] categorize operators into linear oper","cbCaimIIaksTbfof","https://ap.wps.com/l/cbCaimIIaksTbfof","pdf",2379115,1,15,"English","en",105,"# Introduction\n# Privacy-Preserving Machine Learning Background\n# Fixed vs Dynamic Hybrid Protocol Assignment\n# CostSphere Framework and Optimization Method","[{\"question\":\"Why do fixed hybrid protocol assignments hurt privacy-preserving inference performance?\",\"answer\":\"Fixed assignments fail to adapt to changing network conditions, causing communication and latency costs to vary across LAN and WAN environments. This leads to inconsistent runtime efficiency.\"},{\"question\":\"How does CostSphere improve protocol selection for FHE and MPC?\",\"answer\":\"CostSphere uses a predictive cost model and an ILP-based solver to dynamically assign FHE and MPC protocols based on network context, optimizing computational efficiency.\"},{\"question\":\"What is the role of the cost model and ILP solver in CostSphere?\",\"answer\":\"The cost model estimates operator-level costs using MLIR’s TOSA-level dialect, and the ILP solver selects a protocol assignment that minimizes predicted runtime under current network conditions.\"}]","CostSphere - A Cost Model-Driven Privacy-Preserving Machine Learning Framework with Network Context Adaptation | PDF",1785939146,38,{"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},"costsphere-a-cost-model-driven-privacy-preserving-machine-learning-framework-with-network-context-adaptation","",{"@graph":36,"@context":86},[37,54,69],{"@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/costsphere-a-cost-model-driven-privacy-preserving-machine-learning-framework-with-network-context-adaptation/127473/",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","2026-08-23","2026-08-05",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},"Why do fixed hybrid protocol assignments hurt privacy-preserving inference performance?","Question",{"text":76,"@type":77},"Fixed assignments fail to adapt to changing network conditions, causing communication and latency costs to vary across LAN and WAN environments. This leads to inconsistent runtime efficiency.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does CostSphere improve protocol selection for FHE and MPC?",{"text":81,"@type":77},"CostSphere uses a predictive cost model and an ILP-based solver to dynamically assign FHE and MPC protocols based on network context, optimizing computational efficiency.",{"name":83,"@type":74,"acceptedAnswer":84},"What is the role of the cost model and ILP solver in CostSphere?",{"text":85,"@type":77},"The cost model estimates operator-level costs using MLIR’s TOSA-level dialect, and the ILP solver selects a protocol assignment that minimizes predicted runtime under current network conditions.","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},[94,98,102,106,111,116,121,124,129,132,136],{"id":20,"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":112,"doc_module":4,"doc_module_name":46,"category_name":113,"show_sort_weight":114,"slug":115},6,"Technology",50,"technology",{"id":117,"doc_module":4,"doc_module_name":46,"category_name":118,"show_sort_weight":119,"slug":120},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":122,"slug":123},30,"research-report",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":126,"show_sort_weight":127,"slug":128},9,"Religion & Spirituality",20,"religion-spirituality",{"id":127,"doc_module":4,"doc_module_name":46,"category_name":130,"show_sort_weight":127,"slug":131},"World Cup","world-cup",{"id":133,"doc_module":4,"doc_module_name":46,"category_name":134,"show_sort_weight":133,"slug":135},10,"Lifestyle","lifestyle",{"id":137,"doc_module":4,"doc_module_name":46,"category_name":138,"show_sort_weight":107,"slug":139},19,"General","general"]