{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,2,22]],"date-time":"2025-02-22T00:36:02Z","timestamp":1740184562315,"version":"3.37.3"},"reference-count":49,"publisher":"IOP Publishing","issue":"4","license":[{"start":{"date-parts":[[2022,10,28]],"date-time":"2022-10-28T00:00:00Z","timestamp":1666915200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"},{"start":{"date-parts":[[2022,10,28]],"date-time":"2022-10-28T00:00:00Z","timestamp":1666915200000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/iopscience.iop.org\/info\/page\/text-and-data-mining"}],"funder":[{"name":"C-BRIC, one of six centers in JUMP, a Semiconductor Research Corporation (SRC) program sponsored by DARPA"}],"content-domain":{"domain":["iopscience.iop.org"],"crossmark-restriction":false},"short-container-title":["Neuromorph. Comput. Eng."],"published-print":{"date-parts":[[2022,12,1]]},"abstract":"<jats:title>Abstract<\/jats:title>\n               <jats:p>There is an increasing need for continual learning in dynamic systems at the edge, such as self-driving vehicles, surveillance drones, and robotic systems. Such a system requires learning from the data stream, training the model to preserve previous information and adapt to a new task, and generating a single-headed vector for future inference, within a limited power budget. Different from previous continual learning algorithms with dynamic structures, this work focuses on a single network and model segmentation to mitigate catastrophic forgetting problem. Leveraging the redundant capacity of a single network, model parameters for each task are separated into two groups: one important group which is frozen to preserve current knowledge, and a secondary group to be saved (not pruned) for future learning. A fixed-size memory containing a small amount of previously seen data is further adopted to assist the training. Without additional regularization, the simple yet effective approach of progressive segmented training (PST) successfully incorporates multiple tasks and achieves state-of-the-art accuracy in the single-head evaluation on the CIFAR-10 and CIFAR-100 datasets. Moreover, the segmented training significantly improves computation efficiency in continual learning and thus, enabling efficient continual learning at the edge. On Intel Stratix-10 MX FPGA, we further demonstrate the efficiency of PST with representative CNNs trained on CIFAR-10.<\/jats:p>","DOI":"10.1088\/2634-4386\/ac9899","type":"journal-article","created":{"date-parts":[[2022,10,9]],"date-time":"2022-10-09T22:11:20Z","timestamp":1665353480000},"page":"044006","update-policy":"https:\/\/doi.org\/10.1088\/crossmark-policy","source":"Crossref","is-referenced-by-count":1,"title":["Efficient continual learning at the edge with progressive segmented training"],"prefix":"10.1088","volume":"2","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-1079-0347","authenticated-orcid":true,"given":"Xiaocong","family":"Du","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Shreyas Kolala","family":"Venkataramanaiah","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Zheng","family":"Li","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Han-Sok","family":"Suh","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Shihui","family":"Yin","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Gokul","family":"Krishnan","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Frank","family":"Liu","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Jae-sun","family":"Seo","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Yu","family":"Cao","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"266","published-online":{"date-parts":[[2022,10,28]]},"reference":[{"key":"nceac9899bib1","doi-asserted-by":"publisher","first-page":"3521","DOI":"10.1073\/pnas.1611835114","volume":"114","author":"Kirkpatrick","year":"2017","journal-title":"Proc. Natl Acad. Sci. USA"},{"year":"2017","author":"Zenke","key":"nceac9899bib2"},{"year":"2019","author":"Aljundi","key":"nceac9899bib3"},{"year":"2018","author":"Chaudhry","key":"nceac9899bib4"},{"year":"2018","author":"Chaudhry","key":"nceac9899bib5"},{"key":"nceac9899bib6","first-page":"4652","article-title":"Overcoming catastrophic forgetting by incremental moment matching","author":"Lee","year":"2017"},{"key":"nceac9899bib7","first-page":"6467","article-title":"Gradient episodic memory for continual learning","volume":"vol 30","author":"Lopez-Paz","year":"2017"},{"year":"2016","author":"Rebuffi","key":"nceac9899bib8"},{"year":"2019","author":"Zhang","key":"nceac9899bib9"},{"year":"2019","author":"Zhang","key":"nceac9899bib10"},{"key":"nceac9899bib11","first-page":"109","volume":"vol 24","author":"McCloskey","year":"1989"},{"year":"2016","author":"Rusu","key":"nceac9899bib12"},{"key":"nceac9899bib13","first-page":"899","article-title":"Reinforced continual learning","author":"Xu","year":"2018"},{"year":"2017","author":"Yoon","key":"nceac9899bib14"},{"year":"2018","author":"Serr\u00e0","key":"nceac9899bib15"},{"key":"nceac9899bib16","doi-asserted-by":"publisher","first-page":"144","DOI":"10.1007\/978-3-030-01219-9_9","article-title":"Memory aware synapses: learning what (not) to forget","volume":"vol 11207","author":"Aljundi","year":"2018"},{"year":"2016","author":"Kirkpatrick","key":"nceac9899bib17"},{"key":"nceac9899bib18","doi-asserted-by":"publisher","first-page":"2935","DOI":"10.1109\/tpami.2017.2773081","volume":"40","author":"Li","year":"2018","journal-title":"IEEE Trans. Pattern Anal. Mach. Intell."},{"key":"nceac9899bib19","doi-asserted-by":"publisher","first-page":"72","DOI":"10.1007\/978-3-030-01225-0_5","article-title":"Piggyback: adapting a single network to multiple tasks by learning to mask weights","volume":"vol 11208","author":"Mallya","year":"2018"},{"key":"nceac9899bib20","doi-asserted-by":"publisher","first-page":"7765","DOI":"10.1109\/cvpr.2018.00810","article-title":"PackNet: adding multiple tasks to a single network by iterative pruning","author":"Mallya","year":"2018"},{"key":"nceac9899bib21","doi-asserted-by":"publisher","first-page":"233","DOI":"10.1007\/978-3-030-01258-8_15","article-title":"End-to-end incremental learning","author":"Castro","year":"2018"},{"year":"2018","author":"Javed","key":"nceac9899bib22"},{"year":"2018","author":"Farquhar","key":"nceac9899bib23"},{"year":"2017","author":"Nguyen","key":"nceac9899bib24"},{"article-title":"Learning multiple layers of features from tiny images","year":"2009","author":"Krizhevsky","key":"nceac9899bib25"},{"year":"2018","author":"Dhar","key":"nceac9899bib26"},{"year":"2019","author":"Wu","key":"nceac9899bib27"},{"key":"nceac9899bib28","first-page":"5962","article-title":"Memory replay gans: learning to generate new categories without forgetting","author":"Wu","year":"2018"},{"year":"2018","author":"Rios","key":"nceac9899bib29"},{"year":"2019","author":"Wen","key":"nceac9899bib30"},{"year":"2017","author":"Seff","key":"nceac9899bib31"},{"year":"2014","author":"Simonyan","key":"nceac9899bib32"},{"key":"nceac9899bib33","doi-asserted-by":"publisher","first-page":"1629","DOI":"10.48550\/arXiv.1905.11550","article-title":"Single-net continual learning with progressive segmented training","author":"Du","year":"2019"},{"key":"nceac9899bib34","doi-asserted-by":"publisher","first-page":"770","DOI":"10.1109\/cvpr.2016.90","article-title":"Deep residual learning for image recognition","author":"He","year":"2016"},{"year":"2016","author":"Molchanov","key":"nceac9899bib35"},{"key":"nceac9899bib36","first-page":"1135","article-title":"Learning both weights and connections for efficient neural network","author":"Han","year":"2015"},{"year":"2016","author":"Li","key":"nceac9899bib37"},{"key":"nceac9899bib38","first-page":"6869","article-title":"Quantized neural networks: Training neural networks with low precision weights and activations","volume":"18","author":"Hubara","year":"2017","journal-title":"J. Mach. Learn. Res."},{"article-title":"CGaP: continuous growth and pruning for efficient deep learning","year":"2019","author":"Du","key":"nceac9899bib39"},{"year":"2014","author":"Gong","key":"nceac9899bib40"},{"year":"2018","author":"Liu","key":"nceac9899bib41"},{"article-title":"Automatic differentiation in pytorch","year":"2017","author":"Paszke","key":"nceac9899bib42"},{"year":"2015","author":"Ioffe","key":"nceac9899bib43"},{"key":"nceac9899bib44","doi-asserted-by":"publisher","first-page":"5058","DOI":"10.1109\/iccv.2017.541","article-title":"Thinet: a filter level pruning method for deep neural network compression","author":"Luo","year":"2017"},{"key":"nceac9899bib45","first-page":"166","article-title":"Automatic compiler based FPGA accelerator for CNN training","author":"Venkataramanaiah","year":"2019"},{"key":"nceac9899bib46","doi-asserted-by":"publisher","first-page":"1354","DOI":"10.1109\/tvlsi.2018.2815603","volume":"26","author":"Ma","year":"2018","journal-title":"IEEE Trans. Very Large Scale Integr. (VLSI) Syst."},{"key":"nceac9899bib47","doi-asserted-by":"publisher","DOI":"10.1145\/3400302.3415643","article-title":"FPGA-Based low-batch training accelerator for modern CNNs featuring high bandwidth memory","author":"Venkataramanaiah","year":"2020"},{"year":"2016","author":"Deo","key":"nceac9899bib48"},{"key":"nceac9899bib49","doi-asserted-by":"publisher","first-page":"5237","DOI":"10.24963\/ijcai.2020\/755","article-title":"Efficient and modularized training on FPGA for real-time applications","author":"Venkataramanaiah","year":"2020"}],"container-title":["Neuromorphic Computing and Engineering"],"original-title":[],"link":[{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2634-4386\/ac9899","content-type":"text\/html","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2634-4386\/ac9899\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2634-4386\/ac9899","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2634-4386\/ac9899\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2634-4386\/ac9899\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2634-4386\/ac9899\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2634-4386\/ac9899\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"similarity-checking"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2634-4386\/ac9899\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2022,10,27]],"date-time":"2022-10-27T13:46:43Z","timestamp":1666878403000},"score":1,"resource":{"primary":{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2634-4386\/ac9899"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2022,10,28]]},"references-count":49,"journal-issue":{"issue":"4","published-online":{"date-parts":[[2022,10,28]]},"published-print":{"date-parts":[[2022,12,1]]}},"URL":"https:\/\/doi.org\/10.1088\/2634-4386\/ac9899","relation":{},"ISSN":["2634-4386"],"issn-type":[{"type":"electronic","value":"2634-4386"}],"subject":[],"published":{"date-parts":[[2022,10,28]]},"assertion":[{"value":"Efficient continual learning at the edge with progressive segmented training","name":"article_title","label":"Article Title"},{"value":"Neuromorphic Computing and Engineering","name":"journal_title","label":"Journal Title"},{"value":"paper","name":"article_type","label":"Article Type"},{"value":"\u00a9 2022 The Author(s). Published by IOP Publishing Ltd","name":"copyright_information","label":"Copyright Information"},{"value":"2022-06-17","name":"date_received","label":"Date Received","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2022-10-09","name":"date_accepted","label":"Date Accepted","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2022-10-28","name":"date_epub","label":"Online publication date","group":{"name":"publication_dates","label":"Publication dates"}}]}}