[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-128641-en":3,"doc-seo-128641-105":31,"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":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":28,"seo_description":14,"update_tm":29,"read_time":30},128641,962084925636,"Olivia Brown","https://ap-avatar.wpscdn.com/davatar_994ba38a5ba835b3df7d355c54d3ed8d",8,"Research & Report","Towards In-Network Machine Learning on Programmable Network Devices - Doctor of Philosophy Thesis","Machine learning services increasingly run over networks, where rising compute demand and data volume strain traditional server-centric approaches that require higher-performance machines and larger network capacity. In-network computing offers in-band packet processing, early traffic termination, and scalable execution, but applying it to ML inference is a novel challenge because network devices are built for forwarding, not for ML. This thesis presents bottom-up design methods, mapping, and deployment frameworks to enable in-network ML on programmable devices, then addresses accuracy limits via hybrid and distributed strategies, and validates practical use in anomaly detection, IoT traffic classification, financial prediction, and load balancing.","Towards In-Network Machine Learning on Programmable Network Devices  \nChanggang Zheng Jesus College University of Oxford  \nA thesis submitted for the degree of Doctor of Philosophy  \nHilary 2024  \nThis thesis is dedicated to my parents, family, supervisor, colleagues, friends,  \nand all those whose presence illuminates my life.谨以此文献给我的父母，家人，老师，以及所有关心和帮助我的同学和朋友们。  \nABSTRACT  \nMachine learning (ML) services have entered various aspects of our daily lives and industrial production, with most of these services being delivered over networks. The escalating demand for ML services has resulted in increased processing requirements and a growing volume of data. This poses significant challenges to traditional solutions, demanding higher-performance computers and increased network capacity. In-network computing is a promising and scalable solution, capable of processing packets in-band at the pace of data generation and facilitating early termination of traffic. However, using in-network computing for ML inference is a new research avenue. Network devices are designed for high-performance packet processing and forwarding, and their architecture is not intended for ML.  \nThis research addresses the design challenges of in-network ML using a bottom-up approach. It proposes mapping techniques and deployment solutions that overcome existing limitations and enable the Internet to provide in-network ML services on programmable network devices, revolutionising the way we use the Internet for ML. The first part of the thesis focuses on mapping methodologies, which offer efficient mapping solutions for diverse ML models and hardware devices. Then, to support all aforementioned models and mappings, this thesis proposes a rapid prototyping framework, accommodating diverse programmable network devices and use cases. However, while in-network ML can be deployed, its inference accuracy falls short of that achieved by server-based solutions. To address this challenge, the third part of the thesis focuses on a hybrid deployment framework, utilising a large ML model in the backend to assist small in-network models. This hybrid deployment allows close to optimal inference performance while retaining most decisions within the network. The fourth part of the thesis approaches the same problem from a different perspective,  \nproposing a distributed deployment framework. By jointly utilising unused resources among programmable network devices distributed across the network, this solution further scales in-network ML models and computing functions. The last part of the thesis demonstrates the application of in-network ML in anomaly detection, IoT traffic classification, financial market prediction, and load balancing use cases, showing its potential as a practical service.  \nDECLARATION  \nThis thesis is submitted to the Department of Engineering Science, University of Oxford, in fulfilment of the requirements for the degree of Doctor of Philosophy. This work took place at the University of Oxford, with a three-month external visit at the Yale University, United States of America. This thesis is entirely my work, and except where otherwise stated, describes my research.  \nChanggang Zheng Jesus College, Hilary 2024  \nACKNOWLEDGEMENTS  \nThis thesis would not have been possible without the help and support of so many individuals and institutes. Here, I want to express my sincere gratitude.  \nFirst and foremost I would like to thank my advisor, Professor Noa Zilberman, for her trust, guidance, help, and support, which gave me the chance to pursue my doctoral studies at the University of Oxford, do research with the best supervision, and finally have this thesis. Professor Zilberman showed me a rigorous research attitude, set high standards and requirements for me, facilitated my immersion into the research community, and taught me methodologies and approaches for conducting research. I greatly appreciate her mentorship, which helps me learn a lot.  \nI have been fo","cbCaitDPNZ3Vb5PI","https://ap.wps.com/l/cbCaitDPNZ3Vb5PI","pdf",6838330,2,1,234,"English","en",105,"# Abstract\n## Mapping and deployment for in-network ML\n## Rapid prototyping for programmable devices\n## Hybrid inference assistance\n## Distributed deployment using unused network resources\n## Use cases and practical evaluation","[{\"question\":\"Why is in-network computing attractive for machine learning services?\",\"answer\":\"It enables packet processing in-band at the pace of data generation and supports early termination of traffic, improving scalability compared with server-only approaches.\"},{\"question\":\"What are the main technical challenges when deploying ML inference on network devices?\",\"answer\":\"Network devices are optimized for high-performance forwarding and their architecture is not intended for ML workloads, leading to mapping and deployment constraints and reduced inference accuracy.\"},{\"question\":\"How does the thesis improve inference performance for in-network ML?\",\"answer\":\"It proposes a hybrid deployment framework that uses a large model in the backend to assist smaller in-network models, aiming for near-optimal inference while keeping most decisions inside the network.\"}]","Towards In-Network Machine Learning on Programmable Network Devices - Doctor of Philosophy Thesis | PDF",1786002260,590,{"code":4,"msg":32,"data":33},"ok",{"site_id":25,"language":24,"slug":34,"title":13,"keywords":35,"description":14,"schema_data":36,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":29},"towards-in-network-machine-learning-on-programmable-network-devices-doctor-of-philosophy-thesis","",{"@graph":37,"@context":86},[38,54,69],{"@type":39,"itemListElement":40},"BreadcrumbList",[41,45,48,51],{"item":42,"name":43,"@type":44,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":46,"name":47,"@type":44,"position":20},"https://docshare.wps.com/document/","Document",{"item":49,"name":12,"@type":44,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":44,"position":53},"https://docshare.wps.com/document/towards-in-network-machine-learning-on-programmable-network-devices-doctor-of-philosophy-thesis/128641/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":42,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-24","2026-08-06",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 is in-network computing attractive for machine learning services?","Question",{"text":76,"@type":77},"It enables packet processing in-band at the pace of data generation and supports early termination of traffic, improving scalability compared with server-only approaches.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"What are the main technical challenges when deploying ML inference on network devices?",{"text":81,"@type":77},"Network devices are optimized for high-performance forwarding and their architecture is not intended for ML workloads, leading to mapping and deployment constraints and reduced inference accuracy.",{"name":83,"@type":74,"acceptedAnswer":84},"How does the thesis improve inference performance for in-network ML?",{"text":85,"@type":77},"It proposes a hybrid deployment framework that uses a large model in the backend to assist smaller in-network models, aiming for near-optimal inference while keeping most decisions inside the network.","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":25},{"code":4,"msg":5,"data":93},[94,98,102,106,111,116,121,124,129,132,136],{"id":21,"doc_module":4,"doc_module_name":47,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":20,"doc_module":4,"doc_module_name":47,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":47,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":47,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":112,"doc_module":4,"doc_module_name":47,"category_name":113,"show_sort_weight":114,"slug":115},6,"Technology",50,"technology",{"id":117,"doc_module":4,"doc_module_name":47,"category_name":118,"show_sort_weight":119,"slug":120},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":47,"category_name":12,"show_sort_weight":122,"slug":123},30,"research-report",{"id":125,"doc_module":4,"doc_module_name":47,"category_name":126,"show_sort_weight":127,"slug":128},9,"Religion & Spirituality",20,"religion-spirituality",{"id":127,"doc_module":4,"doc_module_name":47,"category_name":130,"show_sort_weight":127,"slug":131},"World Cup","world-cup",{"id":133,"doc_module":4,"doc_module_name":47,"category_name":134,"show_sort_weight":133,"slug":135},10,"Lifestyle","lifestyle",{"id":137,"doc_module":4,"doc_module_name":47,"category_name":138,"show_sort_weight":107,"slug":139},19,"General","general"]