002906008 001__ 2906008
002906008 005__ 20241002102730.0
002906008 0247_ $$2DOI$$9IEEE$$a10.1109/CAMAD59638.2023.10478391$$qpublication
002906008 0248_ $$aoai:cds.cern.ch:2906008$$pcerncds:FULLTEXT$$pcerncds:CERN:FULLTEXT$$pcerncds:CERN
002906008 037__ $$9arXiv$$aarXiv:2310.08087$$ceess.SP
002906008 035__ $$9arXiv$$aoai:arXiv.org:2310.08087
002906008 035__ $$9Inspire$$aoai:inspirehep.net:2774939$$d2024-10-01T15:55:13Z$$h2024-10-02T02:01:01Z$$mmarcxml$$ttrue$$uhttps://inspirehep.net/api/oai2d
002906008 035__ $$9Inspire$$a2774939
002906008 041__ $$aeng
002906008 100__ $$aBarbieri, Luca$$uMilan, Polytech.$$vPolitecnico di Milano, Milan, Italy$$vConsiglio Nazionale delle Ricerche, Milan, Italy
002906008 245__ $$9IEEE$$aA Carbon Tracking Model for Federated Learning: Impact of Quantization and Sparsification
002906008 260__ $$c2023-11-06
002906008 269__ $$c2023-10-12
002906008 300__ $$a6 p
002906008 520__ $$9IEEE$$aFederated Learning (FL) methods adopt efficient communication technologies to distribute machine learning tasks across edge devices, reducing the overhead in terms of data storage and computational complexity compared to centralized solutions. Rather than moving large data volumes from producers (sensors, machines) to energy-hungry data centers, raising environmental concerns due to resource demands, FL provides an alternative solution to mitigate the energy demands of several learning tasks while enabling new Artificial Intelligence of Things (AIoT) applications. This paper proposes a framework for real-time monitoring of the energy and carbon footprint impacts of FL systems. The carbon tracking tool is evaluated for consensus (fully decentralized) and classical FL policies. For the first time, we present a quantitative evaluation of different computationally and communication efficient FL methods from the perspectives of energy consumption and carbon equivalent emissions, suggesting also general guidelines for energy-efficient design. Results indicate that consensus-driven FL implementations should be preferred for limiting carbon emissions when the energy efficiency of the communication is low (i.e., <25Kbit/Joule). Besides, quantization and sparsification operations are shown to strike a balance between learning performances and energy consumption, leading to sustainable FL designs.
002906008 520__ $$9arXiv$$aFederated Learning (FL) methods adopt efficient communication technologies to distribute machine learning tasks across edge devices, reducing the overhead in terms of data storage and computational complexity compared to centralized solutions. Rather than moving large data volumes from producers (sensors, machines) to energy-hungry data centers, raising environmental concerns due to resource demands, FL provides an alternative solution to mitigate the energy demands of several learning tasks while enabling new Artificial Intelligence of Things (AIoT) applications. This paper proposes a framework for real-time monitoring of the energy and carbon footprint impacts of FL systems. The carbon tracking tool is evaluated for consensus (fully decentralized) and classical FL policies. For the first time, we present a quantitative evaluation of different computationally and communication efficient FL methods from the perspectives of energy consumption and carbon equivalent emissions, suggesting also general guidelines for energy-efficient design. Results indicate that consensus-driven FL implementations should be preferred for limiting carbon emissions when the energy efficiency of the communication is low (i.e., < 25 Kbit/Joule). Besides, quantization and sparsification operations are shown to strike a balance between learning performances and energy consumption, leading to sustainable FL designs.
002906008 540__ $$3preprint$$aCC BY 4.0$$uhttp://creativecommons.org/licenses/by/4.0/
002906008 542__ $$3publication$$dIEEE$$g2023
002906008 595__ $$cCDS
002906008 65017 $$2arXiv$$acs.LG
002906008 65017 $$2SzGeCERN$$aComputing and Computers
002906008 65017 $$2arXiv$$aeess.SP
002906008 690C_ $$aCERN
002906008 690C_ $$aARTICLE
002906008 700__ $$aSavazzi, Stefano$$vConsiglio Nazionale delle Ricerche, Milan, Italy
002906008 700__ $$aKianoush, Sanaz$$vConsiglio Nazionale delle Ricerche, Milan, Italy
002906008 700__ $$aNicoli, Monica$$uMilan, Polytech.$$vPolitecnico di Milano, Milan, Italy
002906008 700__ $$aSerio, Luigi$$uCERN$$vTechnology Department, CERN, Geneva 23, Switzerland
002906008 773__ $$c213-218$$y2023
002906008 8564_ $$82549198$$s44361$$uhttps://cds.cern.ch/record/2906008/files/results_carbon_minst.png$$y00001 Analysis of the validation accuracy achieved by FA and CFA schemes under different levels of quantization and sparsification. Rings group together curves related to the same carbon footprint.
002906008 8564_ $$82549199$$s77982$$uhttps://cds.cern.ch/record/2906008/files/CFA_vs_FA.png$$y00000 Federated Averaging (FA) relying on (a) Parameter Server (PS) for model aggregation and (b) Consensus process based on CHOCO-SGD tool. Sparsification and quantization operations and on-device carbon tracking.
002906008 8564_ $$82549200$$s40255$$uhttps://cds.cern.ch/record/2906008/files/results_comm_efficiency.png$$y00002 Analysis of the validation accuracy under carbon constraints with different communication (energy) efficiencies as well as carbon intensities.
002906008 8564_ $$82549201$$s840446$$uhttps://cds.cern.ch/record/2906008/files/2310.08087.pdf$$yFulltext
002906008 960__ $$a13
002906008 962__ $$b2905975$$k213-218$$nedinburgh20231106
002906008 980__ $$aConferencePaper
002906008 980__ $$aARTICLE