{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,5,18]],"date-time":"2026-05-18T01:09:27Z","timestamp":1779066567142,"version":"3.51.4"},"reference-count":39,"publisher":"World Scientific Pub Co Pte Ltd","issue":"10","funder":[{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"crossref","award":["62302221"],"award-info":[{"award-number":["62302221"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"crossref"}]},{"DOI":"10.13039\/501100004608","name":"Natural Science Foundation of Jiangsu Province","doi-asserted-by":"crossref","award":["BK20230913"],"award-info":[{"award-number":["BK20230913"]}],"id":[{"id":"10.13039\/501100004608","id-type":"DOI","asserted-by":"crossref"}]},{"DOI":"10.13039\/501100012226","name":"Fundamental Research Funds for the Central Universities","doi-asserted-by":"crossref","award":["30925010515"],"award-info":[{"award-number":["30925010515"]}],"id":[{"id":"10.13039\/501100012226","id-type":"DOI","asserted-by":"crossref"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["J CIRCUIT SYST COMP"],"published-print":{"date-parts":[[2026,6]]},"abstract":"<jats:p>Deep neural networks (DNNs) are widely used in artificial intelligence (AI) applications due to their powerful representational capacity. However, deploying DNNs on edge devices faces severe challenges, including limited computational resources, restricted memory capacity and heterogeneous processor architectures, which often make conventional DNN models infeasible. In this paper, we propose a novel multi-level model lightweighting framework designed to address these challenges. The framework integrates three key techniques: Neural architecture search (NAS), knowledge distillation (KD) and layer-adaptive quantization. NAS is employed to automatically discover an efficient network architecture under hardware constraints, enabling structural compression. Subsequently, KD transfers knowledge from the original model to the compressed model, mitigating the accuracy loss resulting from compression. Finally, a layer-adaptive quantization strategy is implemented, assigning distinct bit-widths to different layers based on their data distribution characteristics, thereby enhancing storage and computational efficiency without sacrificing performance. Experimental evaluations on edge devices with multiple different processor architectures demonstrate the efficacy of the framework, achieving a model compression rate ranging from 15% to 70% with less than a 1% reduction in accuracy.<\/jats:p>","DOI":"10.1142\/s0218126626500283","type":"journal-article","created":{"date-parts":[[2025,10,24]],"date-time":"2025-10-24T02:59:24Z","timestamp":1761274764000},"source":"Crossref","is-referenced-by-count":1,"title":["A Multi-Level Lightweight Framework for Deep Neural Networks on Resource-Constrained Edge Devices"],"prefix":"10.1142","volume":"35","author":[{"ORCID":"https:\/\/orcid.org\/0009-0008-2717-7566","authenticated-orcid":false,"given":"Jiaying","family":"Yu","sequence":"first","affiliation":[{"name":"School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0009-0001-5561-7994","authenticated-orcid":false,"given":"Haoqiu","family":"Luo","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Xinfa","family":"Dai","sequence":"additional","affiliation":[{"name":"Wuhan Digital Engineering Institute, Wuhan 430205, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0009-0008-7253-0285","authenticated-orcid":false,"given":"Yong","family":"Wu","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0009-0006-3766-4929","authenticated-orcid":false,"given":"Peijin","family":"Cong","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-7734-4077","authenticated-orcid":false,"given":"Junlong","family":"Zhou","sequence":"additional","affiliation":[{"name":"School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"219","published-online":{"date-parts":[[2025,11,25]]},"reference":[{"key":"S0218126626500283BIB001","doi-asserted-by":"publisher","DOI":"10.1007\/s10462-024-10721-6"},{"key":"S0218126626500283BIB002","doi-asserted-by":"crossref","unstructured":"F. Ariai, J. Mackenzie and G. Demartini, Natural language processing for the legal domain: A survey of tasks, datasets, models, and challenges, preprint (2024), arXiv:2410.21306.","DOI":"10.1145\/3777009"},{"key":"S0218126626500283BIB003","doi-asserted-by":"publisher","DOI":"10.1016\/j.inffus.2024.102422"},{"key":"S0218126626500283BIB004","doi-asserted-by":"publisher","DOI":"10.1109\/JIOT.2021.3081422"},{"key":"S0218126626500283BIB005","doi-asserted-by":"publisher","DOI":"10.1109\/TSC.2025.3552338"},{"key":"S0218126626500283BIB006","doi-asserted-by":"publisher","DOI":"10.1109\/TMC.2025.3553201"},{"key":"S0218126626500283BIB007","doi-asserted-by":"publisher","DOI":"10.1109\/TCAD.2024.3513892"},{"key":"S0218126626500283BIB008","doi-asserted-by":"publisher","DOI":"10.1145\/3524500"},{"key":"S0218126626500283BIB009","unstructured":"G. Hinton, O. Vinyals and J. Dean, Distilling the knowledge in a neural network, preprint (2015), arXiv:1503.02531."},{"key":"S0218126626500283BIB010","doi-asserted-by":"publisher","DOI":"10.1145\/3623402"},{"key":"S0218126626500283BIB011","unstructured":"A. Romero, N. Ballas, S. E. Kahou, A. Chassang, C. Gatta and Y. Bengio, Fitnets: Hints for thin deep nets, preprint (2014), arXiv:1412.6550."},{"key":"S0218126626500283BIB012","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2018.00286"},{"key":"S0218126626500283BIB013","first-page":"177","volume-title":"2021 USENIX Annual Technical Conference (USENIX ATC 21)","author":"Zhou Q.","year":"2021"},{"key":"S0218126626500283BIB014","first-page":"6105","volume-title":"Int. Conf. Machine Learning","author":"Tan M.","year":"2019"},{"key":"S0218126626500283BIB015","unstructured":"H. Cai, C. Gan, T. Wang, Z. Zhang and S. Han, Once-for-all: Train one network and specialize it for efficient deployment, preprint (2019), arXiv:1908.09791."},{"key":"S0218126626500283BIB016","unstructured":"H. Liu, K. Simonyan and Y. Yang, Darts: Differentiable architecture search, preprint (2018), arXiv:1806.09055."},{"key":"S0218126626500283BIB017","doi-asserted-by":"publisher","DOI":"10.1109\/TC.2024.3449108"},{"key":"S0218126626500283BIB018","unstructured":"A. Fayyazi, M. Kamal and M. Pedram, MARCO: Hardware-aware neural architecture search for edge devices with multi-agent reinforcement learning and conformal prediction filtering, preprint (2015), arXiv:2506.13755."},{"key":"S0218126626500283BIB019","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2019.00361"},{"key":"S0218126626500283BIB020","first-page":"10347","volume-title":"Int. Conf. Machine Learning","author":"Touvron H.","year":"2021"},{"key":"S0218126626500283BIB021","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.naacl-long.72"},{"key":"S0218126626500283BIB022","first-page":"22379","volume-title":"Forty-second International Conference on Machine Learning","author":"He C.","year":"2025"},{"key":"S0218126626500283BIB023","unstructured":"M. Nagel, M. Fournarakis, R. A. Amjad, Y. Bondarenko, M. Van Baalen and T. Blankevoort, A white paper on neural network quantization, preprint (2021), arXiv:2106.08295."},{"key":"S0218126626500283BIB024","first-page":"7543","volume-title":"Int. Conf. Machine Learning","author":"Zhao R.","year":"2019"},{"key":"S0218126626500283BIB025","first-page":"7197","volume-title":"Int. Conf. Machine Learning","author":"Nagel M.","year":"2020"},{"key":"S0218126626500283BIB026","unstructured":"I. Hubara, Y. Nahshan, Y. Hanani, R. Banner and D. Soudry, Improving post training neural quantization: Layer-wise calibration and integer programming, preprint (2020), arXiv:2006.10518."},{"key":"S0218126626500283BIB027","unstructured":"Y. Li, R. Gong, X. Tan, Y. Yang, P. Hu, Q. Zhang, F. Yu, W. Wang and S. Gu, Brecq: Pushing the limit of post-training quantization by block reconstruction, preprint (2021), arXiv:2102.05426."},{"key":"S0218126626500283BIB028","doi-asserted-by":"crossref","unstructured":"J. Kim, S. Chang and N. Kwak, PQK: Model compression via pruning, quantization, and knowledge distillation, preprint (2021), arXiv:2106.14681.","DOI":"10.21437\/Interspeech.2021-248"},{"key":"S0218126626500283BIB029","doi-asserted-by":"publisher","DOI":"10.1109\/CVPRW63382.2024.00177"},{"key":"S0218126626500283BIB030","doi-asserted-by":"publisher","DOI":"10.1109\/ACCESS.2022.3170561"},{"key":"S0218126626500283BIB031","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"S0218126626500283BIB032","first-page":"84","volume":"60","author":"Krizhevsky A.","year":"2012","journal-title":"Adv. Neural Inf. Process. Syst."},{"key":"S0218126626500283BIB033","doi-asserted-by":"publisher","DOI":"10.1109\/5.726791"},{"key":"S0218126626500283BIB034","doi-asserted-by":"publisher","DOI":"10.1109\/TGRS.2017.2685945"},{"key":"S0218126626500283BIB035","unstructured":"A. Coates, A. Ng and H. Lee, An analysis of single-layer networks in unsupervised feature learning,\n                      Proc. Fourteenth Int. Conf. Artificial Intelligence and Statistics\n                      , JMLR Workshop and Conference Proceedings (PMLR, 2011), pp. 215\u2013223."},{"key":"S0218126626500283BIB036","unstructured":"H. Xiao, K. Rasul and R. Vollgraf, Fashion-MNIST: A novel image dataset for benchmarking machine learning algorithms, preprint (2017), arXiv:1708.07747."},{"key":"S0218126626500283BIB037","first-page":"4","volume-title":"NIPS Workshop on Deep Learning and Unsupervised Feature Learning","author":"Netzer Y."},{"key":"S0218126626500283BIB038","doi-asserted-by":"publisher","DOI":"10.1109\/MSP.2012.2211477"},{"key":"S0218126626500283BIB039","unstructured":"A. Krizhevsky and G. Hinton, Learning multiple layers of features from tiny images (2009), https:\/\/www.cs.toronto.edu\/kriz\/learning-features-2009-TR.pdf."}],"container-title":["Journal of Circuits, Systems and Computers"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.worldscientific.com\/doi\/pdf\/10.1142\/S0218126626500283","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,5,18]],"date-time":"2026-05-18T01:00:52Z","timestamp":1779066052000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.worldscientific.com\/doi\/10.1142\/S0218126626500283"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,11,25]]},"references-count":39,"journal-issue":{"issue":"10","published-print":{"date-parts":[[2026,6]]}},"alternative-id":["10.1142\/S0218126626500283"],"URL":"https:\/\/doi.org\/10.1142\/s0218126626500283","relation":{},"ISSN":["0218-1266","1793-6454"],"issn-type":[{"value":"0218-1266","type":"print"},{"value":"1793-6454","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,11,25]]},"article-number":"2650028"}}