{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CZFYMMBGPHE76TSZLYQ4KCNGMP","short_pith_number":"pith:CZFYMMBG","schema_version":"1.0","canonical_sha256":"164b86302679c9ff4e595e21c509a663d06519848a7f053d9766ba9ab0e2c52c","source":{"kind":"arxiv","id":"2510.01460","version":4},"attestation_state":"computed","paper":{"title":"The Three Regimes of Offline-to-Online Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Lu Li, Pierre-Luc Bacon, Tianwei Ni, Yihao Sun","submitted_at":"2025-10-01T20:58:14Z","abstract_excerpt":"Offline-to-online reinforcement learning (RL) has emerged as a practical paradigm that leverages offline datasets for pretraining and online interactions for fine-tuning. However, its empirical behavior is highly inconsistent: design choices of online fine-tuning that work well in one setting can fail completely in another. Guided by the stability--plasticity principle, we propose a framework that can explain this inconsistency: We argue that efficient fine-tuning must preserve the utility of the stronger offline prior, whether that is the pretrained policy or the offline dataset, while mainta"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2510.01460","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-10-01T20:58:14Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"5358f4eaa605e52c7d4a6f166aa4c6c6e9054b289bb2ef688f08b98ca01cacc5","abstract_canon_sha256":"8c52c0124b1aa2dd7455c67090700914c65295d8d8f44d04bd430435c6a84cc3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-07T02:17:10.836242Z","signature_b64":"YnS00/vM5NqFB80XOEM/7cmOy/wK/u7qNoGy2BsVdycD5EUNI7wD4xTm99WI16upXjHWK1wfBtgRDUxNw3siCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"164b86302679c9ff4e595e21c509a663d06519848a7f053d9766ba9ab0e2c52c","last_reissued_at":"2026-07-07T02:17:10.835346Z","signature_status":"signed_v1","first_computed_at":"2026-07-07T02:17:10.835346Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Three Regimes of Offline-to-Online Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Lu Li, Pierre-Luc Bacon, Tianwei Ni, Yihao Sun","submitted_at":"2025-10-01T20:58:14Z","abstract_excerpt":"Offline-to-online reinforcement learning (RL) has emerged as a practical paradigm that leverages offline datasets for pretraining and online interactions for fine-tuning. However, its empirical behavior is highly inconsistent: design choices of online fine-tuning that work well in one setting can fail completely in another. Guided by the stability--plasticity principle, we propose a framework that can explain this inconsistency: We argue that efficient fine-tuning must preserve the utility of the stronger offline prior, whether that is the pretrained policy or the offline dataset, while mainta"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2510.01460","kind":"arxiv","version":4},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2510.01460/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2510.01460","created_at":"2026-07-07T02:17:10.835457+00:00"},{"alias_kind":"arxiv_version","alias_value":"2510.01460v4","created_at":"2026-07-07T02:17:10.835457+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2510.01460","created_at":"2026-07-07T02:17:10.835457+00:00"},{"alias_kind":"pith_short_12","alias_value":"CZFYMMBGPHE7","created_at":"2026-07-07T02:17:10.835457+00:00"},{"alias_kind":"pith_short_16","alias_value":"CZFYMMBGPHE76TSZ","created_at":"2026-07-07T02:17:10.835457+00:00"},{"alias_kind":"pith_short_8","alias_value":"CZFYMMBG","created_at":"2026-07-07T02:17:10.835457+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2606.25527","citing_title":"Beyond One-Size-Fits-All: Diagnosis-Driven Online Reinforcement Learning with Offline Priors","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2605.14497","citing_title":"ROAD: Adaptive Data Mixing for Offline-to-Online Reinforcement Learning via Bi-Level Optimization","ref_index":16,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CZFYMMBGPHE76TSZLYQ4KCNGMP","json":"https://pith.science/pith/CZFYMMBGPHE76TSZLYQ4KCNGMP.json","graph_json":"https://pith.science/api/pith-number/CZFYMMBGPHE76TSZLYQ4KCNGMP/graph.json","events_json":"https://pith.science/api/pith-number/CZFYMMBGPHE76TSZLYQ4KCNGMP/events.json","paper":"https://pith.science/paper/CZFYMMBG"},"agent_actions":{"view_html":"https://pith.science/pith/CZFYMMBGPHE76TSZLYQ4KCNGMP","download_json":"https://pith.science/pith/CZFYMMBGPHE76TSZLYQ4KCNGMP.json","view_paper":"https://pith.science/paper/CZFYMMBG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2510.01460&json=true","fetch_graph":"https://pith.science/api/pith-number/CZFYMMBGPHE76TSZLYQ4KCNGMP/graph.json","fetch_events":"https://pith.science/api/pith-number/CZFYMMBGPHE76TSZLYQ4KCNGMP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CZFYMMBGPHE76TSZLYQ4KCNGMP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CZFYMMBGPHE76TSZLYQ4KCNGMP/action/storage_attestation","attest_author":"https://pith.science/pith/CZFYMMBGPHE76TSZLYQ4KCNGMP/action/author_attestation","sign_citation":"https://pith.science/pith/CZFYMMBGPHE76TSZLYQ4KCNGMP/action/citation_signature","submit_replication":"https://pith.science/pith/CZFYMMBGPHE76TSZLYQ4KCNGMP/action/replication_record"}},"created_at":"2026-07-07T02:17:10.835457+00:00","updated_at":"2026-07-07T02:17:10.835457+00:00"}