[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-120740-en":3,"doc-seo-120740-105":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":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},120740,549758146520,"Patrick","https://ap-avatar.wpscdn.com/avatar/80002397d8c0411e94?_k=1775819394049821470",8,"Research & Report","Towards Energy-Aware Machine Learning in Geo-Distributed IoT Settings - Conference paper","IoT increasingly brings machine learning to network edge environments, but the energy cost of computationally intensive training directly drives carbon emissions. The work argues for integrating energy awareness into ML pipeline decision-making, specifically for when and where to initiate training in geo-distributed IoT services. Using a traffic-monitoring use case with mobile edge computing, the discussion emphasizes grid carbon-intensity volatility and quantifies carbon reductions by scheduling training across different times and locations.","Towards Energy-Aware Machine Learning in Geo-Distributed IoT Settings  \nDemetris Trihinas 1[0000−0002−9540−7342] and Lauritz Thamsen2[0000−0003−3755−1503]  \n1 Department of Computer Science, University of Nicosia  \n[trihinas.d@unic.ac.cy](trihinas.d@unic.ac.cy)  \n2 School of Computing Science, University of Glasgow  \n[lauritz.thamsen@glasgow.ac.uk](lauritz.thamsen@glasgow.ac.uk)  \nAbstract. As the Internet of Things (IoT) increasingly empowers the network extremes with in-place intelligence through Machine Learning (ML), energy consumption and carbon emissions become crucial factors.  \nML is often computationally intensive, with state-of-the-art model architectures consuming significant energy per training round and imposing a large carbon footprint. This work, therefore, argues for the need to introduce novel mechanisms into the ML pipelines of IoT services, so that energy awareness is integrated in the decision-making process for when and where to initiate ML model training.  \nKeywords: Machine Learning · Internet of Things · Distributed Systems · Energy Profiling · Carbon Footprint · System Orchestration  \n1 Introduction  \nWith recent advancements in IoT hardware, we are seeing the use of ML on IoT devices for highly responsive and intelligent services. However, ML is computehungry. In fact, the computational power required for training new state-ofthe-art model architectures has been doubling every 4 months [2] . This computational effort results in higher and higher energy consumption and, in turn, increasing carbon emissions, contributing to global warming. Already, ICT organizations report that approximately 15% of their energy consumption can be attributed to AI/ML and this ratio is expected to rise considerably [3] . With Gartner [1] indicating that 75% of enterprise data will be created and processed outside of data centers, and the climate crisis demanding a rapid reduction in carbon emissions, a key emerging challenge is to adequately support the migration to sustainable AI-driven cloud edge IoT solutions [5] .  \nThis work discusses the challenges of deploying AI-driven IoT services in geo-distributed settings with a focus on energy consumption and carbon footprint. During the session we will broaden the discussion towards the need for extending ML orchestration frameworks so that their decision-making mechanisms cover energy-awareness by recommending when and where ML models should be trained and elaborate why these two inter-related challenges are not easy to overcome.  \n2 D. Trihinas & L. Thamsen  \nFig. 1. Cyprus 24h carbon intensity  \nFig. 2. Cyprus 24h power production  \nFig. 3. Sweden 24h carbon intensity  \n2 Reference Use Case  \nTo drive the discussion, let us consider a realistic ML-driven IoT application. This application features several road-side IoT units, using cameras and object detection for traffic monitoring. Several Mobile Edge Computing nodes (MECs) are scattered across the city and employed for local coordination as well as recurrent model training at a neighborhood level. For the evaluation we consider a MEC to be powered by a DELL PowerEdge R610 server and equipped with a Nvidia T4 GPU. The ML pipeline employs the TensorFlow benchmark suite3 to output a CNN model for object detection, trained with the ImageNet dataset4 (144GB, 1.3M images) for a duration of approximately 5 hours, when it reaches a satisfactory MLPerf accuracy.  \n3 When to Train a ML Model?  \nDeciding when to initiate repeated ML model training can highly impact the carbon footprint of an ML-based application. In particular, an application’s operational carbon footprint depends on the energy mix powering the compute resources used. An illustrative example is given in Fig. 1, where for a given day in the country of Cyprus, the carbon intensity of the energy grid shows significant volatility. This is attributed to the mix of energy sources powering the grid (Fig. 2), where the low-carbon energy sources solar and wind generate to ","cbCailLm43mL5YtP","https://ap.wps.com/l/cbCailLm43mL5YtP","pdf",394718,1,5,"English","en",105,"# Introduction\n## Reference Use Case\n## When to Train a ML Model?\n## Where to Train a Model?\n## Energy-Aware Support for ML Workflow Orchestration","[{\"question\":\"Why does energy awareness matter for ML in geo-distributed IoT settings?\",\"answer\":\"ML training is computationally intensive and increases energy consumption, which in turn raises carbon emissions. Energy awareness helps reduce the environmental footprint while deploying ML in IoT services across regions.\"},{\"question\":\"How can timing affect the carbon footprint of model training?\",\"answer\":\"When training is initiated depends on the grid’s carbon intensity, which can vary significantly during the day. The document shows that starting mid-day instead of evening hours can noticeably reduce carbon emissions.\"},{\"question\":\"Why does the training location influence environmental impact?\",\"answer\":\"Different countries or regions rely on different energy mixes, producing different carbon intensity profiles. Migrating training to lower-carbon locations can substantially reduce emissions even for comparable training schedules.\"}]","Towards Energy-Aware Machine Learning in Geo-Distributed IoT Settings - Conference paper | PDF",1785731788,13,{"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},"towards-energy-aware-machine-learning-in-geo-distributed-iot-settings-conference-paper","",{"@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/towards-energy-aware-machine-learning-in-geo-distributed-iot-settings-conference-paper/120740/",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":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"Why does energy awareness matter for ML in geo-distributed IoT settings?","Question",{"text":75,"@type":76},"ML training is computationally intensive and increases energy consumption, which in turn raises carbon emissions. Energy awareness helps reduce the environmental footprint while deploying ML in IoT services across regions.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How can timing affect the carbon footprint of model training?",{"text":80,"@type":76},"When training is initiated depends on the grid’s carbon intensity, which can vary significantly during the day. The document shows that starting mid-day instead of evening hours can noticeably reduce carbon emissions.",{"name":82,"@type":73,"acceptedAnswer":83},"Why does the training location influence environmental impact?",{"text":84,"@type":76},"Different countries or regions rely on different energy mixes, producing different carbon intensity profiles. Migrating training to lower-carbon locations can substantially reduce emissions even for comparable training schedules.","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},[93,97,101,105,109,114,119,122,127,130,134],{"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":21,"doc_module":4,"doc_module_name":46,"category_name":106,"show_sort_weight":107,"slug":108},"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":46,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":46,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":46,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":46,"category_name":128,"show_sort_weight":125,"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":21,"slug":137},19,"General","general"]