{"date_updated":"2024-10-09T21:05:40Z","ec_funded":1,"article_processing_charge":"No","_id":"11458","citation":{"chicago":"Krumes, Alexandra, Eugenia B Iofinova, Adrian Vladu, and Dan-Adrian Alistarh. “AC/DC: Alternating Compressed/DeCompressed Training of Deep Neural Networks.” In 35th Conference on Neural Information Processing Systems, 34:8557–70. Curran Associates, 2021.","ista":"Krumes A, Iofinova EB, Vladu A, Alistarh D-A. 2021. AC/DC: Alternating Compressed/DeCompressed training of deep neural networks. 35th Conference on Neural Information Processing Systems. NeurIPS: Neural Information Processing Systems vol. 34, 8557–8570.","apa":"Krumes, A., Iofinova, E. B., Vladu, A., & Alistarh, D.-A. (2021). AC/DC: Alternating Compressed/DeCompressed training of deep neural networks. In 35th Conference on Neural Information Processing Systems (Vol. 34, pp. 8557–8570). Virtual, Online: Curran Associates.","ama":"Krumes A, Iofinova EB, Vladu A, Alistarh D-A. AC/DC: Alternating Compressed/DeCompressed training of deep neural networks. In: 35th Conference on Neural Information Processing Systems. Vol 34. Curran Associates; 2021:8557-8570.","ieee":"A. Krumes, E. B. Iofinova, A. Vladu, and D.-A. Alistarh, “AC/DC: Alternating Compressed/DeCompressed training of deep neural networks,” in 35th Conference on Neural Information Processing Systems, Virtual, Online, 2021, vol. 34, pp. 8557–8570.","short":"A. Krumes, E.B. Iofinova, A. Vladu, D.-A. Alistarh, in:, 35th Conference on Neural Information Processing Systems, Curran Associates, 2021, pp. 8557–8570.","mla":"Krumes, Alexandra, et al. “AC/DC: Alternating Compressed/DeCompressed Training of Deep Neural Networks.” 35th Conference on Neural Information Processing Systems, vol. 34, Curran Associates, 2021, pp. 8557–70."},"intvolume":" 34","scopus_import":"1","publication_status":"published","author":[{"last_name":"Peste","first_name":"Elena-Alexandra","id":"32D78294-F248-11E8-B48F-1D18A9856A87","full_name":"Peste, Elena-Alexandra"},{"id":"f9a17499-f6e0-11ea-865d-fdf9a3f77117","first_name":"Eugenia B","last_name":"Iofinova","orcid":"0000-0002-7778-3221","full_name":"Iofinova, Eugenia B"},{"full_name":"Vladu, Adrian","first_name":"Adrian","last_name":"Vladu"},{"id":"4A899BFC-F248-11E8-B48F-1D18A9856A87","first_name":"Dan-Adrian","last_name":"Alistarh","orcid":"0000-0003-3650-940X","full_name":"Alistarh, Dan-Adrian"}],"date_published":"2021-12-06T00:00:00Z","conference":{"name":"NeurIPS: Neural Information Processing Systems","start_date":"2021-12-06","end_date":"2021-12-14","location":"Virtual, Online"},"related_material":{"record":[{"status":"public","relation":"dissertation_contains","id":"13074"}]},"oa":1,"arxiv":1,"corr_author":"1","status":"public","type":"conference","publication":"35th Conference on Neural Information Processing Systems","oa_version":"Published Version","quality_controlled":"1","page":"8557-8570","month":"12","department":[{"_id":"GradSch"},{"_id":"DaAl"}],"abstract":[{"text":"The increasing computational requirements of deep neural networks (DNNs) have led to significant interest in obtaining DNN models that are sparse, yet accurate. Recent work has investigated the even harder case of sparse training, where the DNN weights are, for as much as possible, already sparse to reduce computational costs during training. Existing sparse training methods are often empirical and can have lower accuracy relative to the dense baseline. In this paper, we present a general approach called Alternating Compressed/DeCompressed (AC/DC) training of DNNs, demonstrate convergence for a variant of the algorithm, and show that AC/DC outperforms existing sparse training methods in accuracy at similar computational budgets; at high sparsity levels, AC/DC even outperforms existing methods that rely on accurate pre-trained dense models. An important property of AC/DC is that it allows co-training of dense and sparse models, yielding accurate sparse–dense model pairs at the end of the training process. This is useful in practice, where compressed variants may be desirable for deployment in resource-constrained settings without re-doing the entire training flow, and also provides us with insights into the accuracy gap between dense and compressed models. The code is available at: https://github.com/IST-DASLab/ACDC.","lang":"eng"}],"date_created":"2022-06-20T12:11:53Z","publisher":"Curran Associates","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","publication_identifier":{"isbn":["9781713845393"],"issn":["1049-5258"]},"day":"6","external_id":{"arxiv":["2106.12379"]},"acknowledged_ssus":[{"_id":"ScienComp"}],"project":[{"call_identifier":"H2020","grant_number":"805223","name":"Elastic Coordination for Scalable Machine Learning","_id":"268A44D6-B435-11E9-9278-68D0E5697425"}],"volume":34,"year":"2021","title":"AC/DC: Alternating Compressed/DeCompressed training of deep neural networks","main_file_link":[{"url":"https://proceedings.neurips.cc/paper/2021/file/48000647b315f6f00f913caa757a70b3-Paper.pdf","open_access":"1"}],"acknowledgement":"This project has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 805223 ScaleML), and a CNRS PEPS grant. This research was supported by the Scientific Service Units (SSU) of IST Austria through resources provided by Scientific Computing (SciComp). We would also like to thank Christoph Lampert for his feedback on an earlier version of this work, as well as for providing hardware for the Transformer-XL experiments.","language":[{"iso":"eng"}]}