{"paper":{"title":"EMA: Efficient Model Adaptation for Learning-based Systems","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"EMA lets learning-based systems adapt to changing environments by aligning new states to past ones and prioritizing useful data labels, cutting retraining costs.","cross_cats":["cs.DC","cs.NI"],"primary_cat":"cs.LG","authors_text":"Daiyang Yu, Fan Lai, Xinyu Chen, Yan Liang, Yaqi Qiao, Yihan Zhang","submitted_at":"2026-05-13T17:26:06Z","abstract_excerpt":"Machine learning (ML) is increasingly applied to optimize system performance in tasks such as resource management and network simulation. Unlike traditional ML tasks (e.g., image classification), networked systems often operate in heterogeneous, long-running, and dynamic environment states, where input conditions (e.g., network loads) and operational objectives can shift over time and across settings. Existing learning-based systems offer little support for adaptation, resulting in costly model training, extensive data collection, degraded system performance, and slow responsiveness.\n  This pa"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"This paper presents EMA, the first model adaptation system supporting learning-based systems to adapt to evolving environments with minimal operational overhead... Evaluations on eight representative learning-based systems show that EMA reduces adaptation costs (e.g., GPU training time) by 14.9-42.4% while improving system performance (e.g., network throughput) by 6.9-31.3%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That state transformers can reliably map new environment inputs to similar prior states across heterogeneous system designs, and that the utility-based labeling prioritization balances training and labeling costs without missing critical decision data in dynamic settings.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"EMA cuts adaptation costs in learning-based systems by 14.9-42.4% and raises performance by 6.9-31.3% via state transformers for input alignment and prioritized high-utility data labeling.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"EMA lets learning-based systems adapt to changing environments by aligning new states to past ones and prioritizing useful data labels, cutting retraining costs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"4b4606780d66b27933ad06991642b10176440688889208ecc3ca5eb39a099ec6"},"source":{"id":"2605.13942","kind":"arxiv","version":1},"verdict":{"id":"49bf7f2f-efb0-46e8-90d7-39edf158f29a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T06:14:30.849439Z","strongest_claim":"This paper presents EMA, the first model adaptation system supporting learning-based systems to adapt to evolving environments with minimal operational overhead... Evaluations on eight representative learning-based systems show that EMA reduces adaptation costs (e.g., GPU training time) by 14.9-42.4% while improving system performance (e.g., network throughput) by 6.9-31.3%.","one_line_summary":"EMA cuts adaptation costs in learning-based systems by 14.9-42.4% and raises performance by 6.9-31.3% via state transformers for input alignment and prioritized high-utility data labeling.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That state transformers can reliably map new environment inputs to similar prior states across heterogeneous system designs, and that the utility-based labeling prioritization balances training and labeling costs without missing critical decision data in dynamic settings.","pith_extraction_headline":"EMA lets learning-based systems adapt to changing environments by aligning new states to past ones and prioritizing useful data labels, cutting retraining costs."},"references":{"count":43,"sample":[{"doi":"","year":2016,"title":"Brendan McMahan, Ilya Mironov, Kunal Talwar, and Li Zhang","work_id":"f5727c1d-a12c-41ea-8aa0-3a4626c4c4a3","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Venkat Arun and Hari Balakrishnan. 2018. Copa: Practical Delay-Based Congestion Control for the Internet. InNSDI","work_id":"5b967dfc-6c1e-4c36-8d60-a9d9c50bec3a","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Simon Eismann, Long Bui, Johannes Grohmann, Cristina Abad, Nikolas Herbst, and Samuel Kounev. 2021. Sizeless: Predicting the optimal size of serverless functions. InMiddleware. 248–259","work_id":"9dc208b4-0070-4028-acf9-547474cc8240","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Xianghong Fang, Haoli Bai, Ziyi Guo, Bin Shen, Steven Hoi, and Zenglin Xu. 2020. DART: Domain-adversarial residual-transfer net- works for unsupervised cross-domain image classification.Neural Network","work_id":"22bd3bfd-fcf8-4e78-a525-b654182c919d","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Matthew Honnibal, Ines Montani, Sofie Van Lan- deghem, and Adriane Boyd","work_id":"97beb0d3-662a-47d1-acaa-7ee761ae92d3","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":43,"snapshot_sha256":"4f4e600a4111007ef97d60825ab0b3e2080416411db37b0e09dbd6f23ea2068f","internal_anchors":1},"formal_canon":{"evidence_count":2,"snapshot_sha256":"f2fe4d05ade7f711c32689d9c225f9b4af3f53c3e6dc75be15798e2e4983b707"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}