{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2025,9,23]],"date-time":"2025-09-23T02:14:22Z","timestamp":1758593662674,"version":"3.44.0"},"reference-count":43,"publisher":"Oxford University Press (OUP)","issue":"9","license":[{"start":{"date-parts":[[2025,6,9]],"date-time":"2025-06-09T00:00:00Z","timestamp":1749427200000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/academic.oup.com\/pages\/standard-publication-reuse-rights"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2025,9,21]]},"abstract":"<jats:title>Abstract<\/jats:title>\n               <jats:p>Machine unlearning in the context of cybersecurity and privacy protection facilitates the removal of specific training data impacts from deep learning (DL) models, adhering to security, privacy, or compliance demands. However, traditional methods can only handle short-term, independent unlearning tasks. Conversely, real-world scenarios often involve extensive unlearning demands from users. Current methods fail to adequately address these demands due to substantial computational overhead and adverse impacts on inference accuracy, leaving the security and privacy of many users at risk. To navigate these challenges adeptly, we introduce the Multi-Agent Reinforcement Learning Data Lifecycle Management (MADLM) strategy. MADLM intricately examines the interactions between unlearning and continuous learning processes, enabling the postponement of certain tasks for combined execution to optimize computational resources. Concurrently, it employs strategic data management to maintain and enhance inference accuracy. Furthermore, by utilizing Multi-Agent Reinforcement Learning (MARL), MADLM dynamically orchestrates task scheduling to minimize computational demands, improve task response times, and bolster inference reliability, crucial for upholding stringent cybersecurity and privacy standards. Our evaluations of MADLM reveal substantial enhancements, including a 6% uplift in inference accuracy and a dramatic reduction in computational overhead to merely 12% of the original demands, effectively expanding the data security protections.<\/jats:p>","DOI":"10.1093\/comjnl\/bxaf031","type":"journal-article","created":{"date-parts":[[2025,3,15]],"date-time":"2025-03-15T08:16:32Z","timestamp":1742026592000},"page":"1197-1207","source":"Crossref","is-referenced-by-count":0,"title":["Efficient unlearning for data security in deep learning systems"],"prefix":"10.1093","volume":"68","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-3462-1505","authenticated-orcid":false,"given":"Enting","family":"Guo","sequence":"first","affiliation":[{"name":"Department of Computer and Information Systems , University of Aizu, 90 Kami-iawase, Tsuruga, Ikki-machi, Aizu-Wakamatsu, Fukushima, 965-0006,","place":["Japan"]}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-6461-9684","authenticated-orcid":false,"given":"Chunhua","family":"Su","sequence":"additional","affiliation":[{"name":"Department of Computer and Information Systems , University of Aizu, 90 Kami-iawase, Tsuruga, Ikki-machi, Aizu-Wakamatsu, Fukushima, 965-0006,","place":["Japan"]}],"role":[{"role":"author","vocabulary":"crossref"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-5303-0700","authenticated-orcid":false,"given":"Peng","family":"Li","sequence":"additional","affiliation":[{"name":"Department of Computer and Information Systems , University of Aizu, 90 Kami-iawase, Tsuruga, Ikki-machi, Aizu-Wakamatsu, Fukushima, 965-0006,","place":["Japan"]}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"286","published-online":{"date-parts":[[2025,6,9]]},"reference":[{"key":"2025092201571189100_ref1","first-page":"419","article-title":"Learning to communicate effectively between battery-free devices","volume-title":"Proceedings of NSDI USENIX Association 2022, Renton, WA USA, April 4\u20136","author":"Geissdoerfer","year":"2022"},{"key":"2025092201571189100_ref2","first-page":"1907","article-title":"Miris: Fast object track queries in video","volume-title":"Proceedings of SIGMOD ACM 2020, Portland, Oregon, USA, June 14\u201319","author":"Bastani","year":"2020"},{"author":"General Data Protection Regulation (GDPR)","key":"2025092201571189100_ref3"},{"author":"California Consumer Privacy Act (CCPA)","key":"2025092201571189100_ref4"},{"author":"Protection and Electronic Documents Act (PIPEDA)","key":"2025092201571189100_ref5"},{"key":"2025092201571189100_ref6","first-page":"622","article-title":"Federated unlearning via class-discriminative pruning","volume-title":"Proceedings of WWW ACM 2022, Lyon, France, April 25\u201329","author":"Wang","year":"2022"},{"key":"2025092201571189100_ref7","first-page":"4007","article-title":"On the necessity of auditable algorithmic definitions for machine unlearning","volume-title":"Proceedings of USENIX Security USENIX Association 2022, Boston, MA, USA, August 10\u201312","author":"Thudi","year":"2022"},{"key":"2025092201571189100_ref8","first-page":"896","article-title":"When machine unlearning jeopardizes privacy","author":"Chen","journal-title":"Proceedings of CCS ACM 2021, Austin; TX, USA, New York, NY, USA, ACM"},{"key":"2025092201571189100_ref9","first-page":"7766","article-title":"Boundary unlearning: Rapid forgetting of deep networks via shifting the decision boundary","volume-title":"Proceedings of CVPR IEEE\/CVF 2023, Vancouver, Canada, June 18\u201322","author":"Chen","year":"2023"},{"author":"Bourtoule","journal-title":"Proceedings of S&P IEEE 2021, Virtual Conference","article-title":"Machine unlearning","key":"2025092201571189100_ref10"},{"volume-title":"Proceedings of ECCV Springer 2022","author":"","article-title":"Learning with recoverable forgetting","key":"2025092201571189100_ref11"},{"volume-title":"Proceedings of CVPR IEEE\/CVF 2022","author":"","article-title":"Deep unlearning via randomized conditionally independent hessians","key":"2025092201571189100_ref12"},{"volume-title":"Proceedings of WWW ACM 2022","author":"","article-title":"Recommendation unlearning","key":"2025092201571189100_ref13"},{"volume-title":"Proceedings of S&P IEEE 2022","author":"","article-title":"Unrolling SGD: Understanding factors influencing machine unlearning","key":"2025092201571189100_ref14"},{"volume-title":"Proceedings of CCS ACM 2022","author":"","article-title":"Graph unlearning","key":"2025092201571189100_ref15"},{"volume-title":"Proceedings of NSDI USENIX 2022","author":"","article-title":"Check-N-run: A checkpointing system for training deep learning recommendation models","key":"2025092201571189100_ref16"},{"author":"","journal-title":"Proceedings of CVPR IEEE 2021","article-title":"Semantic-aware knowledge distillation for few-shot class-incremental learning","key":"2025092201571189100_ref17"},{"volume-title":"Proceedings of NSDI USENIX 2022","author":"","article-title":"Ekya: Continuous learning of video analytics models on edge compute servers","key":"2025092201571189100_ref18"},{"volume-title":"Proceedings of NSDI USENIX 2022","author":"","article-title":"Privid: Practical, privacy-preserving video analytics queries","key":"2025092201571189100_ref19"},{"author":"","journal-title":"Proceedings of NeurIPS NeurIPS 2022","article-title":"Multi-agent reinforcement learning is a sequence modeling problem","key":"2025092201571189100_ref20"},{"key":"2025092201571189100_ref21","doi-asserted-by":"publisher","first-page":"1240","DOI":"10.1109\/COMST.2022.3160697","article-title":"Applications of multi-agent reinforcement learning in future internet: A comprehensive survey","volume":"24","author":"Li","year":"2022","journal-title":"IEEE Commun Surv Tutor"},{"key":"2025092201571189100_ref22","doi-asserted-by":"publisher","first-page":"2776","DOI":"10.1109\/TPAMI.2017.2769047","article-title":"One-pass learning with incremental and decremental features","volume":"40","author":"Hou","year":"2017","journal-title":"IEEE Trans Pattern Anal Mach Intell"},{"author":"","journal-title":"Proceedings of ICML PMLR 2020","article-title":"Learning with feature and distribution evolvable streams","key":"2025092201571189100_ref23"},{"author":"","journal-title":"Proceedings of AAAI AAAI 2021","article-title":"Few-shot lifelong learning","key":"2025092201571189100_ref24"},{"key":"2025092201571189100_ref25","first-page":"14771","article-title":"A simple but strong baseline for online continual learning: Repeated augmented rehearsal","volume":"35","author":"Zhang","year":"2022","journal-title":"Advances in Neural Information Processing Systems"},{"volume-title":"Proceedings of AAAI AAAI 2017","author":"","article-title":"Incrementally learning the hierarchical softmax function for neural language models","key":"2025092201571189100_ref26"},{"key":"2025092201571189100_ref27","doi-asserted-by":"publisher","first-page":"3521","DOI":"10.1073\/pnas.1611835114","article-title":"Overcoming catastrophic forgetting in neural networks","volume":"114","author":"Kirkpatrick","year":"2017","journal-title":"Proceedings of NAS 2017"},{"volume-title":"Proceedings of NIPS Curran 2017","author":"","article-title":"Gradient episodic memory for continual learning","key":"2025092201571189100_ref28"},{"volume-title":"Proceedings of CVPR IEEE 2022","author":"","article-title":"Learning to prompt for continual learning","key":"2025092201571189100_ref29"},{"volume-title":"Proceedings of ECCV Springer 2018","author":"","article-title":"Memory aware synapses: Learning what (not) to forget","key":"2025092201571189100_ref30"},{"key":"2025092201571189100_ref31","doi-asserted-by":"publisher","first-page":"2935","DOI":"10.1109\/TPAMI.2017.2773081","article-title":"Learning without forgetting","volume":"40","author":"Li","year":"2017","journal-title":"IEEE Trans Pattern Anal Mach Intell"},{"volume-title":"Proceedings of CVPR IEEE 2020","author":"","article-title":"Eternal sunshine of the spotless net: Selective forgetting in deep networks","key":"2025092201571189100_ref32"},{"volume-title":"Proceedings of AAAI AAAI Press 2024","author":"","article-title":"Fast machine unlearning without retraining through selective synaptic dampening","key":"2025092201571189100_ref33"},{"volume-title":"Proceedings of AAAI AAAI Press 2023","author":"","article-title":"Can bad teaching induce forgetting? Unlearning in deep networks using an incompetent teacher","key":"2025092201571189100_ref34"},{"key":"2025092201571189100_ref35","doi-asserted-by":"publisher","first-page":"31","DOI":"10.1007\/PL00009249","article-title":"Adaptive disk spindown via optimal rent-to-buy in probabilistic environments","volume":"23","author":"Krishnan","year":"1999","journal-title":"Algorithmica"},{"key":"2025092201571189100_ref36","doi-asserted-by":"publisher","first-page":"542","DOI":"10.1007\/BF01189993","article-title":"Competitive randomized algorithms for nonuniform problems","volume":"11","author":"Karlin","year":"1994","journal-title":"Algorithmica"},{"key":"2025092201571189100_ref37","doi-asserted-by":"publisher","first-page":"243","DOI":"10.1145\/375827.375843","article-title":"On-line analysis of the TCP acknowledgment delay problem","volume":"48","author":"Dooly","year":"2001","journal-title":"J ACM"},{"volume-title":"Proceedings of AAAI AAAI Press 2022","author":"","article-title":"A multi-agent reinforcement learning approach for efficient client selection in federated learning","key":"2025092201571189100_ref38"},{"volume-title":"Proceedings of WWW ACM 2023","author":"","article-title":"Learning cooperative oversubscription for cloud by chance-constrained multi-agent reinforcement learning","key":"2025092201571189100_ref39"},{"volume-title":"Proceedings of CVPR IEEE 2016","author":"","article-title":"Deep residual learning for image recognition","key":"2025092201571189100_ref40"},{"author":"","key":"2025092201571189100_ref41"},{"author":"","key":"2025092201571189100_ref42"},{"author":"","key":"2025092201571189100_ref43"}],"container-title":["The Computer Journal"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/academic.oup.com\/comjnl\/article-pdf\/68\/9\/1197\/63458505\/bxaf031.pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/academic.oup.com\/comjnl\/article-pdf\/68\/9\/1197\/63458505\/bxaf031.pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,9,22]],"date-time":"2025-09-22T05:57:23Z","timestamp":1758520643000},"score":1,"resource":{"primary":{"URL":"https:\/\/academic.oup.com\/comjnl\/article\/68\/9\/1197\/8159132"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,6,9]]},"references-count":43,"journal-issue":{"issue":"9","published-online":{"date-parts":[[2025,6,9]]},"published-print":{"date-parts":[[2025,9,21]]}},"URL":"https:\/\/doi.org\/10.1093\/comjnl\/bxaf031","relation":{},"ISSN":["0010-4620","1460-2067"],"issn-type":[{"type":"print","value":"0010-4620"},{"type":"electronic","value":"1460-2067"}],"subject":[],"published-other":{"date-parts":[[2025,9]]},"published":{"date-parts":[[2025,6,9]]}}}