{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:3W7OS5CBNYZQIEXVYDM2XADOXF","short_pith_number":"pith:3W7OS5CB","schema_version":"1.0","canonical_sha256":"ddbee974416e330412f5c0d9ab806eb9417a0dc3b6ddad55bbb0bd2b32113879","source":{"kind":"arxiv","id":"2511.03828","version":3},"attestation_state":"computed","paper":{"title":"From Static Constraints to Dynamic Adaptation: Sample-Level Constraint Relaxation for Offline-to-Online Reinforcement Learning","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"DARE releases constraints at the sample level in offline-to-online reinforcement learning by measuring behavioral consistency instead of data origin.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Lipeng Zu, Shayok Chakraborty, Xiaonan Zhang, Yu Qian","submitted_at":"2025-11-05T19:48:46Z","abstract_excerpt":"Offline-to-online reinforcement learning (O2O RL) faces a central challenge between retaining offline conservatism and adapting to online feedback under distribution shift. This challenge arises because data behavior evolves during fine-tuning, rendering data origin a misleading basis for constraint handling and thereby leading to objective-data mismatch. We therefore propose Dynamic Alignment for RElaxation (DARE), a distribution-aware framework for sample-level constraint relaxation based on the behavioral consistency with a behavior model. To our knowledge, DARE is the first to condition co"},"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":true},"canonical_record":{"source":{"id":"2511.03828","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2025-11-05T19:48:46Z","cross_cats_sorted":[],"title_canon_sha256":"dd3d9dcc035538b0b3095ad4b70d089724d1ad024cac9a35889a31fa43b4eab5","abstract_canon_sha256":"48e281b45390cf9e2137120ac0477b41d9a54c8e7d583129e20fbb2562b7b2de"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:04:16.799604Z","signature_b64":"kibYWNNxMTaVZ2nYOQ7GZsN5yXhg3Ad5uUZwUd7ZAEAmEvw4G1XJtQ+9ow/LZp6lB6l+IN+EtXacYBTPm5OuDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ddbee974416e330412f5c0d9ab806eb9417a0dc3b6ddad55bbb0bd2b32113879","last_reissued_at":"2026-05-20T00:04:16.798543Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:04:16.798543Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From Static Constraints to Dynamic Adaptation: Sample-Level Constraint Relaxation for Offline-to-Online Reinforcement Learning","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"DARE releases constraints at the sample level in offline-to-online reinforcement learning by measuring behavioral consistency instead of data origin.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Lipeng Zu, Shayok Chakraborty, Xiaonan Zhang, Yu Qian","submitted_at":"2025-11-05T19:48:46Z","abstract_excerpt":"Offline-to-online reinforcement learning (O2O RL) faces a central challenge between retaining offline conservatism and adapting to online feedback under distribution shift. This challenge arises because data behavior evolves during fine-tuning, rendering data origin a misleading basis for constraint handling and thereby leading to objective-data mismatch. We therefore propose Dynamic Alignment for RElaxation (DARE), a distribution-aware framework for sample-level constraint relaxation based on the behavioral consistency with a behavior model. To our knowledge, DARE is the first to condition co"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We provide a theoretical analysis showing that behavior-based sample exchange consistently improves the distinction between offline-like and online-like subsets. DARE is the first to condition constraint release on behavioral consistency via a posterior-induced exchange mechanism, moving beyond a binary offline/online data distinction.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that a learned behavior model can reliably produce a posterior that accurately measures per-sample behavioral consistency for the exchange mechanism, and that this consistency metric remains meaningful as the policy evolves during fine-tuning (abstract, paragraph on DARE framework).","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DARE provides a distribution-aware sample-level constraint release mechanism for offline-to-online RL based on behavioral consistency with a behavior model, supported by theoretical analysis and D4RL experiments showing improved stability and performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"DARE releases constraints at the sample level in offline-to-online reinforcement learning by measuring behavioral consistency instead of data origin.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"bd1a0d42a69636c9e08005d7b586a5f6672e8301c9ef1e7cb74897e30c641afb"},"source":{"id":"2511.03828","kind":"arxiv","version":3},"verdict":{"id":"e07e4a47-0996-4507-88b4-cb081da7b491","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-18T00:33:05.695345Z","strongest_claim":"We provide a theoretical analysis showing that behavior-based sample exchange consistently improves the distinction between offline-like and online-like subsets. DARE is the first to condition constraint release on behavioral consistency via a posterior-induced exchange mechanism, moving beyond a binary offline/online data distinction.","one_line_summary":"DARE provides a distribution-aware sample-level constraint release mechanism for offline-to-online RL based on behavioral consistency with a behavior model, supported by theoretical analysis and D4RL experiments showing improved stability and performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that a learned behavior model can reliably produce a posterior that accurately measures per-sample behavioral consistency for the exchange mechanism, and that this consistency metric remains meaningful as the policy evolves during fine-tuning (abstract, paragraph on DARE framework).","pith_extraction_headline":"DARE releases constraints at the sample level in offline-to-online reinforcement learning by measuring behavioral consistency instead of data origin."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2511.03828/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":2,"snapshot_sha256":"ab8ee580afabd57cad2ca9161f612484cdaa8e5636be86c7a2e60fafcecdbb50"},"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":"2511.03828","created_at":"2026-05-20T00:04:16.798672+00:00"},{"alias_kind":"arxiv_version","alias_value":"2511.03828v3","created_at":"2026-05-20T00:04:16.798672+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.03828","created_at":"2026-05-20T00:04:16.798672+00:00"},{"alias_kind":"pith_short_12","alias_value":"3W7OS5CBNYZQ","created_at":"2026-05-20T00:04:16.798672+00:00"},{"alias_kind":"pith_short_16","alias_value":"3W7OS5CBNYZQIEXV","created_at":"2026-05-20T00:04:16.798672+00:00"},{"alias_kind":"pith_short_8","alias_value":"3W7OS5CB","created_at":"2026-05-20T00:04:16.798672+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3W7OS5CBNYZQIEXVYDM2XADOXF","json":"https://pith.science/pith/3W7OS5CBNYZQIEXVYDM2XADOXF.json","graph_json":"https://pith.science/api/pith-number/3W7OS5CBNYZQIEXVYDM2XADOXF/graph.json","events_json":"https://pith.science/api/pith-number/3W7OS5CBNYZQIEXVYDM2XADOXF/events.json","paper":"https://pith.science/paper/3W7OS5CB"},"agent_actions":{"view_html":"https://pith.science/pith/3W7OS5CBNYZQIEXVYDM2XADOXF","download_json":"https://pith.science/pith/3W7OS5CBNYZQIEXVYDM2XADOXF.json","view_paper":"https://pith.science/paper/3W7OS5CB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2511.03828&json=true","fetch_graph":"https://pith.science/api/pith-number/3W7OS5CBNYZQIEXVYDM2XADOXF/graph.json","fetch_events":"https://pith.science/api/pith-number/3W7OS5CBNYZQIEXVYDM2XADOXF/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3W7OS5CBNYZQIEXVYDM2XADOXF/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3W7OS5CBNYZQIEXVYDM2XADOXF/action/storage_attestation","attest_author":"https://pith.science/pith/3W7OS5CBNYZQIEXVYDM2XADOXF/action/author_attestation","sign_citation":"https://pith.science/pith/3W7OS5CBNYZQIEXVYDM2XADOXF/action/citation_signature","submit_replication":"https://pith.science/pith/3W7OS5CBNYZQIEXVYDM2XADOXF/action/replication_record"}},"created_at":"2026-05-20T00:04:16.798672+00:00","updated_at":"2026-05-20T00:04:16.798672+00:00"}