{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/distributed-deep-learning-on-edge-devices","title":"Distributed deep learning on edge-devices: feasibility via adaptive compression","arxiv_id":"1702.04683","date":"2017-02-15","proceeding":null,"authors":["Corentin Hardy","Erwan Le Merrer","Bruno Sericola"],"abstract":"A large portion of data mining and analytic services use modern machine\nlearning techniques, such as deep learning. The state-of-the-art results by\ndeep learning come at the price of an intensive use of computing resources. The\nleading frameworks (e.g., TensorFlow) are executed on GPUs or on high-end\nservers in datacenters. On the other end, there is a proliferation of personal\ndevices with possibly free CPU cycles; this can enable services to run in\nusers' homes, embedding machine learning operations. In this paper, we ask the\nfollowing question: Is distributed deep learning computation on WAN connected\ndevices feasible, in spite of the traffic caused by learning tasks? We show\nthat such a setup rises some important challenges, most notably the ingress\ntraffic that the servers hosting the up-to-date model have to sustain.\n  In order to reduce this stress, we propose adaComp, a novel algorithm for\ncompressing worker updates to the model on the server. Applicable to stochastic\ngradient descent based approaches, it combines efficient gradient selection and\nlearning rate modulation. We then experiment and measure the impact of\ncompression, device heterogeneity and reliability on the accuracy of learned\nmodels, with an emulator platform that embeds TensorFlow into Linux containers.\nWe report a reduction of the total amount of data sent by workers to the server\nby two order of magnitude (e.g., 191-fold reduction for a convolutional network\non the MNIST dataset), when compared to a standard asynchronous stochastic\ngradient descent, while preserving model accuracy.","url_abs":"http://arxiv.org/abs/1702.04683v2","url_pdf":"http://arxiv.org/pdf/1702.04683v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"distributed-deep-learning-on-edge-devices","repo_url":"https://github.com/Hardy-c/AdaComp","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[{"task_slug":"machine-learning","task_name":"BIG-bench Machine Learning"},{"task_slug":null,"task_name":"CPU"},{"task_slug":"deep-learning","task_name":"Deep Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1702.04683","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}