{"tmp":{"image":"/images/cc_by.png","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode"},"date_updated":"2023-06-19T10:44:38Z","article_processing_charge":"No","month":"07","day":"01","file":[{"content_type":"application/pdf","creator":"dernst","access_level":"open_access","file_name":"2021_PMLR_Alimisis.pdf","checksum":"7ec0d59bac268b49c76bf2e036dedd7a","relation":"main_file","success":1,"file_id":"13154","date_created":"2023-06-19T10:41:05Z","file_size":429087,"date_updated":"2023-06-19T10:41:05Z"}],"type":"conference","abstract":[{"lang":"eng","text":"We investigate fast and communication-efficient algorithms for the classic problem of minimizing a sum of strongly convex and smooth functions that are distributed among n\r\n different nodes, which can communicate using a limited number of bits. Most previous communication-efficient approaches for this problem are limited to first-order optimization, and therefore have \\emph{linear} dependence on the condition number in their communication complexity. We show that this dependence is not inherent: communication-efficient methods can in fact have sublinear dependence on the condition number. For this, we design and analyze the first communication-efficient distributed variants of preconditioned gradient descent for Generalized Linear Models, and for Newton’s method. Our results rely on a new technique for quantizing both the preconditioner and the descent direction at each step of the algorithms, while controlling their convergence rate. We also validate our findings experimentally, showing faster convergence and reduced communication relative to previous methods."}],"department":[{"_id":"DaAl"}],"publication_status":"published","date_published":"2021-07-01T00:00:00Z","scopus_import":"1","publisher":"ML Research Press","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","date_created":"2023-06-18T22:00:48Z","oa":1,"oa_version":"Published Version","status":"public","title":"Communication-efficient distributed optimization with quantized preconditioners","quality_controlled":"1","publication_identifier":{"eissn":["2640-3498"],"isbn":["9781713845065"]},"external_id":{"arxiv":["2102.07214"]},"page":"196-206","author":[{"full_name":"Alimisis, Foivos","last_name":"Alimisis","first_name":"Foivos"},{"orcid":"0000-0002-5646-9524","full_name":"Davies, Peter","id":"11396234-BB50-11E9-B24C-90FCE5697425","first_name":"Peter","last_name":"Davies"},{"orcid":"0000-0003-3650-940X","last_name":"Alistarh","first_name":"Dan-Adrian","full_name":"Alistarh, Dan-Adrian","id":"4A899BFC-F248-11E8-B48F-1D18A9856A87"}],"publication":"Proceedings of the 38th International Conference on Machine Learning","has_accepted_license":"1","language":[{"iso":"eng"}],"_id":"13147","ec_funded":1,"conference":{"start_date":"2021-07-18","name":"International Conference on Machine Learning","location":"Virtual","end_date":"2021-07-24"},"citation":{"ieee":"F. Alimisis, P. Davies, and D.-A. Alistarh, “Communication-efficient distributed optimization with quantized preconditioners,” in <i>Proceedings of the 38th International Conference on Machine Learning</i>, Virtual, 2021, vol. 139, pp. 196–206.","mla":"Alimisis, Foivos, et al. “Communication-Efficient Distributed Optimization with Quantized Preconditioners.” <i>Proceedings of the 38th International Conference on Machine Learning</i>, vol. 139, ML Research Press, 2021, pp. 196–206.","short":"F. Alimisis, P. Davies, D.-A. Alistarh, in:, Proceedings of the 38th International Conference on Machine Learning, ML Research Press, 2021, pp. 196–206.","ama":"Alimisis F, Davies P, Alistarh D-A. Communication-efficient distributed optimization with quantized preconditioners. In: <i>Proceedings of the 38th International Conference on Machine Learning</i>. Vol 139. ML Research Press; 2021:196-206.","apa":"Alimisis, F., Davies, P., &#38; Alistarh, D.-A. (2021). Communication-efficient distributed optimization with quantized preconditioners. In <i>Proceedings of the 38th International Conference on Machine Learning</i> (Vol. 139, pp. 196–206). Virtual: ML Research Press.","chicago":"Alimisis, Foivos, Peter Davies, and Dan-Adrian Alistarh. “Communication-Efficient Distributed Optimization with Quantized Preconditioners.” In <i>Proceedings of the 38th International Conference on Machine Learning</i>, 139:196–206. ML Research Press, 2021.","ista":"Alimisis F, Davies P, Alistarh D-A. 2021. Communication-efficient distributed optimization with quantized preconditioners. Proceedings of the 38th International Conference on Machine Learning. International Conference on Machine Learning vol. 139, 196–206."},"year":"2021","volume":139,"intvolume":"       139","ddc":["000"],"project":[{"call_identifier":"H2020","grant_number":"805223","name":"Elastic Coordination for Scalable Machine Learning","_id":"268A44D6-B435-11E9-9278-68D0E5697425"},{"grant_number":"754411","_id":"260C2330-B435-11E9-9278-68D0E5697425","name":"ISTplus - Postdoctoral Fellowships","call_identifier":"H2020"}],"file_date_updated":"2023-06-19T10:41:05Z","acknowledgement":"The authors would like to thank Janne Korhonen, Aurelien Lucchi, Celestine MendlerDunner and Antonio Orvieto for helpful discussions. FA ¨and DA were supported during this work by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 805223 ScaleML). PD was supported by the European Union’s Horizon 2020 programme under the Marie Skłodowska-Curie grant agreement No. 754411."}