[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-125895-en":3,"doc-seo-125895-105":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},125895,2336474466712,"Maeve","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","Accelerating Machine Learning for Trading Using Programmable Switches","High-frequency trading relies on specialized hardware for fast execution but may use simpler decision logic, while machine-learning-driven trading can improve predictions yet suffers from latency that undermines its advantage. This paper proposes an in-network ML approach that embeds feature engineering, feature extraction, and ML inference into programmable network devices. A solution is built for stock mid-price and volatility forecasting using commodity switches, achieving microsecond-scale ultra-low latency and 64%–97% latency reduction versus prior work, while maintaining server-level ML performance. A hybrid deployment strategy keeps misclassification change under 0.8% relative to the server baseline while processing 49% of traffic on-switch and reducing end-to-end latency by 45% on average.","ECAI 2024  \nU. Endriss et al. (Eds.)  \n© 2024 The Authors.  \nThis article is published online with Open Access by IOS Press and distributed under the terms of the Creative Commons Attribution Non-Commercial License 4.0 (CC BY-NC 4.0). doi:10.3233/FAIA240894  \n3429  \nAccelerating Machine Learning for Trading Using  \nProgrammable Switches  \nXinpeng Honga, * , Changgang Zhenga, Stefan Zohrena and Noa ZilbermanaaDepartment of Engineering Science, University of Oxford, Oxford, United Kingdom  \nAbstract. High-frequency trading (HFT) employs cutting-edge hardware for rapid decision-making and order execution but often relies on simpler algorithms that may miss deeper market trends. Conversely, lower-frequency algorithmic trading uses machine learning (ML) for better market predictions but higher latency can negate its strategic benefits. To achieve the best of both worlds, we present an in-network ML solution that embeds ML processes into programmable network devices, accelerating feature engineering and extraction as well as ML inference. In this paper, we design and develop a solution that supports both stock mid-price and volatility movement forecasting using commodity switches. Our approach achieves microsecond-scale, ultra-low latency, significantly lowering it by 64% to 97% compared to previous works, while upholding the same level of ML performance as server models. Additionally, by combining network hardware and servers, a hybrid deployment strategy can keep the misclassification rate change below 0.8% relative to the server baseline while processing 49% ofthe traffic directly on the switch and achieving a 45% average reduction in end-to-end latency.  \n1 Introduction  \nHigh-frequency trading (HFT), a form of algorithmic trading that distinguishes itself by extremely high speeds and microsecond-scale data processing, often leverages specialized hardware for acceleration [22] . In contrast, trading strategies that operate at lower frequencies typically rely on machine learning (ML) to improve decisionmaking processes by analyzing market conditions more accurately and thoroughly [19] . While combining the analytical and predictive power of ML with the rapid execution capabilities of HFT proves tobe highly effective [24], merging ML-based strategic trading with HFT often leads to increased latency and a slowdown in trading speeds [11] . Recognizing the complementary strengths and weaknesses of HFT and ML-driven trading, this study focuses on bridging the gap between these two paradigms.  \nApplication workloads can be partially or fully offloaded from end hosts to the network infrastructure, using programmable network devices that are already deployed for traffic forwarding. This is referred to as in-network computing [47] . Within this domain, in-network ML focuses on deploying pre-trained ML models directly within networking hardware to perform inference. The primary objectives of this strategy include achieving lower latency, higher throughput, and improved power efficiency [39, 49] . Figure 1 illustrates the difference in data paths between general ML and in-network ML. Functioning within a programmable switch, in-network ML can reduce  \n∗ Corresponding Author. Email: [xinpeng.hong@eng.ox.ac.uk](xinpeng.hong@eng.ox.ac.uk).  \n\n| 􀀂􀀃􀀄􀀅􀀆 \u003Cbr>\u003Cbr>\u003Cbr>\u003Cbr>􀀇􀀈􀀃􀀈􀀉􀀊􀀋 􀀌􀀊􀀍􀀎􀀏􀀃􀀈 􀀑􀀈􀀊􀀉􀀃􀀏􀀃􀀒\u003Cbr> |  | \u003Cbr>􀀂􀀃􀀔􀀃􀀈􀀆􀀕􀀖􀀉􀀗 􀀌􀀊􀀍􀀎􀀏􀀃􀀈 􀀑􀀈􀀊􀀉􀀃 􀀏 􀀃􀀒\u003Cbr>\u003Cbr>\u003Cbr> |\n| --- | --- | --- |\n\nFigure 1: Data path difference between general ML and in-network ML.  \nlatency by circumventing both end hosts and intermediate network devices between end hosts and the in-network ML node. Therefore, by design, in-network ML offers an effective solution for accelerating ML-driven applications in scenarios where latency is crucial, especially in trading activities.  \nAdditionally, concerns have risen about the escalating burden on CPU cycles and excessive electrical power use driven by the competition among HFT firms [42] . Beyond mere acceleration, the inherent energy efficiency of networ","cbCaihfN1r2oozmy","https://ap.wps.com/l/cbCaihfN1r2oozmy","pdf",1503397,6,1,"English","en",105,"# Introduction\n## High-frequency trading vs ML-driven trading\n## In-network computing and in-network ML objectives\n## Motivation: latency and energy efficiency\n## Related work and research gap\n## Contributions and evaluation scope","[{\"question\":\"What problem does the paper address in combining ML and high-frequency trading?\",\"answer\":\"It addresses the increased latency and slowed trading speeds that often occur when ML-based strategic decision-making is combined with high-frequency trading workflows.\"},{\"question\":\"How does the proposed in-network ML system differ from traditional datacenter inference?\",\"answer\":\"It embeds feature engineering, feature extraction, and ML inference inside programmable switch hardware, bypassing end-host and intermediate device processing paths to reduce latency.\"},{\"question\":\"What trading tasks and datasets are evaluated in the paper?\",\"answer\":\"The approach is evaluated on stock mid-price movement prediction and, for the first time, short-term stock volatility forecasting using realistic datasets on programmable switches with multiple in-network ML models.\"}]","Accelerating Machine Learning for Trading Using Programmable Switches | PDF",1785901892,20,{"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},"accelerating-machine-learning-for-trading-using-programmable-switches","",{"@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/accelerating-machine-learning-for-trading-using-programmable-switches/125895/",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","2026-08-22","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},"What problem does the paper address in combining ML and high-frequency trading?","Question",{"text":76,"@type":77},"It addresses the increased latency and slowed trading speeds that often occur when ML-based strategic decision-making is combined with high-frequency trading workflows.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does the proposed in-network ML system differ from traditional datacenter inference?",{"text":81,"@type":77},"It embeds feature engineering, feature extraction, and ML inference inside programmable switch hardware, bypassing end-host and intermediate device processing paths to reduce latency.",{"name":83,"@type":74,"acceptedAnswer":84},"What trading tasks and datasets are evaluated in the paper?",{"text":85,"@type":77},"The approach is evaluated on stock mid-price movement prediction and, for the first time, short-term stock volatility forecasting using realistic datasets on programmable switches with multiple in-network ML models.","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"]