{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,23]],"date-time":"2026-10-23T20:16:22Z","timestamp":1792786582042,"version":"4.5.0"},"reference-count":58,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","license":[{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"name":"National Natural Science Foundation of China","award":["62472249"],"award-info":[{"award-number":["62472249"]}],"id":[{"id":"https:\/\/ror.org\/01h0zpd94","id-type":"ROR","asserted-by":"crossref"}]},{"name":"Shenzhen Science and Technology Program","award":["KJZD20240903102300001"],"award-info":[{"award-number":["KJZD20240903102300001"]}]},{"name":"Shenzhen Science and Technology Program","award":["JCYJ20250604145014018"],"award-info":[{"award-number":["JCYJ20250604145014018"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE J. Sel. Areas Commun."],"published-print":{"date-parts":[[2026]]},"DOI":"10.1109\/jsac.2026.3735988","type":"journal-article","created":{"date-parts":[[2026,9,23]],"date-time":"2026-09-23T19:13:21Z","timestamp":1790190801000},"page":"5792-5807","source":"Crossref","is-referenced-by-count":1,"title":["Taming Latency and Bandwidth: A Theoretical Framework and Adaptive Algorithm for Communication-Constrained Training"],"prefix":"10.1109","volume":"44","author":[{"ORCID":"https:\/\/orcid.org\/0009-0007-8356-3026","authenticated-orcid":false,"given":"Rongwei","family":"Lu","sequence":"first","affiliation":[{"name":"Tsinghua University","place":["Shenzhen, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0003-9393-6931","authenticated-orcid":false,"given":"Chunyang","family":"Li","sequence":"additional","affiliation":[{"name":"Tsinghua University","place":["Shenzhen, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-4897-2645","authenticated-orcid":false,"given":"Jingyan","family":"Jiang","sequence":"additional","affiliation":[{"name":"Tsinghua University","place":["Shenzhen, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xingguang","family":"Wei","sequence":"additional","affiliation":[{"name":"University of Science and Technology of China","place":["Hefei, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-5462-6178","authenticated-orcid":false,"given":"Zhi","family":"Wang","sequence":"additional","affiliation":[{"name":"Tsinghua University","place":["Shenzhen, China"]}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","article-title":"GPT-4 technical report","volume-title":"arXiv:2303.08774","author":"Achiam","year":"2023"},{"key":"ref2","article-title":"DeepSeek-V3 technical report","volume-title":"arXiv:2412.19437","author":"Liu","year":"2024"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1073\/pnas.2311878121"},{"key":"ref4","volume-title":"NVLink |NVIDIA Data Center","year":"2023"},{"key":"ref5","article-title":"Data centers in the AI era: Energy and emissions impacts in the U.S. and key states","author":"Springer","year":"2025"},{"key":"ref6","volume-title":"AEP Ohio: Data Center Electricity Demands to Put Central Ohio on Same Tier as Manhattan","author":"Williams","year":"2024"},{"key":"ref7","volume-title":"Deeplink Connects 1,500 Km of Compute Clusters to Train a 100b-Parameter Model","year":"2025"},{"key":"ref8","article-title":"OpenDiLoCo: An open-source framework for globally distributed low-communication training","author":"Jaghouar","year":"2024","journal-title":"arXiv:2407.07852"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1145\/3651890.3672228"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2025.emnlp-main.1486"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP43922.2022.9747113"},{"key":"ref12","article-title":"Language models are unsupervised multitask learners","volume-title":"OpenAI Blog","author":"Radford","year":"2019"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.1109\/JSAC.2022.3192050"},{"key":"ref14","first-page":"29995","article-title":"Delayed gradient averaging: Tolerate the communication latency for federated learning","volume-title":"Proc. NeurIPS","author":"Zhu"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.1109\/JSAC.2021.3087262"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.1109\/JSAC.2025.3559136"},{"key":"ref17","first-page":"14236","article-title":"PowerSGD: Practical low-rank gradient compression for distributed optimization","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"32","author":"Vogels"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.1109\/TMC.2024.3504284"},{"key":"ref19","first-page":"1","article-title":"The error-feedback framework: Better rates for SGD with delayed gradients and compressed updates","volume":"21","author":"Stich","year":"2020","journal-title":"J. Mach. Learn. Res."},{"key":"ref20","first-page":"4384","article-title":"Ef21: A new, simpler, theoretically better, and practically faster error feedback","volume-title":"Proc. NeurIPS","author":"Richt\u00e1rik"},{"key":"ref21","article-title":"EControl: Fast distributed optimization with compression and error control","volume-title":"Proc. 12th Int. Conf. Learn. Represent.","author":"Gao"},{"key":"ref22","first-page":"4452","article-title":"Sparsified SGD with memory","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","volume":"31","author":"Stich"},{"key":"ref23","doi-asserted-by":"publisher","DOI":"10.1109\/INFOCOM48880.2022.9796982"},{"key":"ref24","article-title":"On communication compression for distributed optimization on heterogeneous data","author":"Stich","year":"2020","journal-title":"arXiv:2009.02388"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1109\/jsac.2026.3735988"},{"key":"ref26","article-title":"Deep gradient compression: Reducing the communication bandwidth for distributed training","volume-title":"Proc. ICLR","author":"Lin"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1109\/ICDCS51616.2021.00060"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.52202\/068431-1251"},{"key":"ref29","article-title":"GPipe: Efficient training of giant neural networks using pipeline parallelism","author":"Huang","year":"2018","journal-title":"arXiv:1811.06965"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1145\/3341301.3359646"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1145\/948205.948211"},{"key":"ref32","doi-asserted-by":"publisher","DOI":"10.1145\/1030194.1015470"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.1109\/INFCOM.2002.1019258"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2009.5206848"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.48550\/arXiv.2010.11929"},{"key":"ref37","article-title":"Pointer sentinel mixture models","volume-title":"Proc. Int. Conf. Learn. Represent.","author":"Merity"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.1109\/TNNLS.2019.2944481"},{"key":"ref39","first-page":"55","article-title":"Adaptive gradient communication via critical learning regime identification","volume":"3","author":"Agarwal","year":"2021","journal-title":"Mach. Learn. Syst."},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.1109\/INFOCOM42981.2021.9488810"},{"key":"ref41","first-page":"36058","article-title":"CocktailSGD: Fine-tuning foundation models over 500mbps networks","volume-title":"Proc. ICML","volume":"202","author":"Wang"},{"key":"ref42","first-page":"4387","article-title":"The non-IID data quagmire of decentralized machine learning","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Hsieh"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.1109\/ICDE53745.2022.00077"},{"key":"ref44","first-page":"1","article-title":"$\\gamma $\n-FedHT: Stepsize-aware hard-threshold gradient compression in federated learning","volume-title":"Proc. IEEE INFOCOM - IEEE Conf. Comput. Commun.","author":"Lu"},{"key":"ref45","article-title":"DiLoCo: Distributed low-communication training of language models","author":"Douillard","year":"2023","journal-title":"arXiv:2311.08105"},{"key":"ref46","article-title":"DAGC: Data-volume-aware adaptive sparsification gradient compression for distributed machine learning in mobile computing","author":"Lu","year":"2023","journal-title":"arXiv:2311.07324"},{"key":"ref47","doi-asserted-by":"publisher","DOI":"10.1109\/INFOCOM53939.2023.10229029"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.24963\/ijcai.2024\/458"},{"key":"ref49","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v38i15.29603"},{"key":"ref50","doi-asserted-by":"publisher","DOI":"10.24963\/ijcai.2024\/584"},{"key":"ref51","doi-asserted-by":"publisher","DOI":"10.52202\/079017-0123"},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.1145\/3721146.3721952"},{"key":"ref53","article-title":"CO2: Efficient distributed training with full communication-computation overlap","volume-title":"Proc. ICLR","author":"Sun"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.52202\/068431-1846"},{"key":"ref55","first-page":"29416","article-title":"SWARM parallelism: Training large models can be surprisingly communication-efficient","volume-title":"Proc. ICML","volume":"202","author":"Ryabinin"},{"key":"ref56","article-title":"The llama 3 herd of models","author":"Grattafiori","year":"2024","journal-title":"arXiv:2407.21783"},{"key":"ref57","first-page":"563","article-title":"MAST: Global scheduling of ML training across geo-distributed datacenters at hyperscale","volume-title":"Proc. OSDI","author":"Choudhury"},{"key":"ref58","doi-asserted-by":"publisher","DOI":"10.52202\/085713-3559"}],"container-title":["IEEE Journal on Selected Areas in Communications"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/49\/11372474\/11701246.pdf?arnumber=11701246","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,9,30]],"date-time":"2026-09-30T19:14:35Z","timestamp":1790795675000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11701246\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":58,"URL":"https:\/\/doi.org\/10.1109\/jsac.2026.3735988","relation":{},"ISSN":["1558-0008","0733-8716"],"issn-type":[{"value":"1558-0008","type":"electronic"},{"value":"0733-8716","type":"print"}],"subject":[],"published":{"date-parts":[[2026]]}}}