{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:YFCCBSQE2VLRA7E6OP65O2UZ3T","short_pith_number":"pith:YFCCBSQE","schema_version":"1.0","canonical_sha256":"c14420ca04d557107c9e73fdd76a99dcc4fb83995f6149e595f6377defd4c1fb","source":{"kind":"arxiv","id":"1910.03732","version":1},"attestation_state":"computed","paper":{"title":"Ctrl-Z: Recovering from Instability in Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.RO","stat.ML"],"primary_cat":"cs.LG","authors_text":"Jake Bruce, J\\\"urgen Leitner, Thierry Peynot, Vibhavari Dasagi","submitted_at":"2019-10-09T00:45:53Z","abstract_excerpt":"When learning behavior, training data is often generated by the learner itself; this can result in unstable training dynamics, and this problem has particularly important applications in safety-sensitive real-world control tasks such as robotics. In this work, we propose a principled and model-agnostic approach to mitigate the issue of unstable learning dynamics by maintaining a history of a reinforcement learning agent over the course of training, and reverting to the parameters of a previous agent whenever performance significantly decreases. We develop techniques for evaluating this perform"},"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":"1910.03732","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2019-10-09T00:45:53Z","cross_cats_sorted":["cs.RO","stat.ML"],"title_canon_sha256":"611eff132183e03af5ce898a90b03e728badb303a55ed00f3f328f8701d7766a","abstract_canon_sha256":"cd6ba1b33431b30dfbd860563d9af206c8d0a0e3c8f18403b6261c969e255796"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:10:57.376210Z","signature_b64":"rsFrd7eKow3hHsDXYalODJEzhql6x4vsy8H27e+LdbQHA6auFbC6sBhh1EbjaGNz7Y7Gwx9Q2IgiyQC8/a/aCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c14420ca04d557107c9e73fdd76a99dcc4fb83995f6149e595f6377defd4c1fb","last_reissued_at":"2026-07-05T00:10:57.375766Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:10:57.375766Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Ctrl-Z: Recovering from Instability in Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.RO","stat.ML"],"primary_cat":"cs.LG","authors_text":"Jake Bruce, J\\\"urgen Leitner, Thierry Peynot, Vibhavari Dasagi","submitted_at":"2019-10-09T00:45:53Z","abstract_excerpt":"When learning behavior, training data is often generated by the learner itself; this can result in unstable training dynamics, and this problem has particularly important applications in safety-sensitive real-world control tasks such as robotics. In this work, we propose a principled and model-agnostic approach to mitigate the issue of unstable learning dynamics by maintaining a history of a reinforcement learning agent over the course of training, and reverting to the parameters of a previous agent whenever performance significantly decreases. We develop techniques for evaluating this perform"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1910.03732","kind":"arxiv","version":1},"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/1910.03732/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":"1910.03732","created_at":"2026-07-05T00:10:57.375845+00:00"},{"alias_kind":"arxiv_version","alias_value":"1910.03732v1","created_at":"2026-07-05T00:10:57.375845+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1910.03732","created_at":"2026-07-05T00:10:57.375845+00:00"},{"alias_kind":"pith_short_12","alias_value":"YFCCBSQE2VLR","created_at":"2026-07-05T00:10:57.375845+00:00"},{"alias_kind":"pith_short_16","alias_value":"YFCCBSQE2VLRA7E6","created_at":"2026-07-05T00:10:57.375845+00:00"},{"alias_kind":"pith_short_8","alias_value":"YFCCBSQE","created_at":"2026-07-05T00:10:57.375845+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2412.10381","citing_title":"Supervised Learning-enhanced Multi-Group Actor Critic for Live Stream Allocation in Feed","ref_index":8,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YFCCBSQE2VLRA7E6OP65O2UZ3T","json":"https://pith.science/pith/YFCCBSQE2VLRA7E6OP65O2UZ3T.json","graph_json":"https://pith.science/api/pith-number/YFCCBSQE2VLRA7E6OP65O2UZ3T/graph.json","events_json":"https://pith.science/api/pith-number/YFCCBSQE2VLRA7E6OP65O2UZ3T/events.json","paper":"https://pith.science/paper/YFCCBSQE"},"agent_actions":{"view_html":"https://pith.science/pith/YFCCBSQE2VLRA7E6OP65O2UZ3T","download_json":"https://pith.science/pith/YFCCBSQE2VLRA7E6OP65O2UZ3T.json","view_paper":"https://pith.science/paper/YFCCBSQE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1910.03732&json=true","fetch_graph":"https://pith.science/api/pith-number/YFCCBSQE2VLRA7E6OP65O2UZ3T/graph.json","fetch_events":"https://pith.science/api/pith-number/YFCCBSQE2VLRA7E6OP65O2UZ3T/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YFCCBSQE2VLRA7E6OP65O2UZ3T/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YFCCBSQE2VLRA7E6OP65O2UZ3T/action/storage_attestation","attest_author":"https://pith.science/pith/YFCCBSQE2VLRA7E6OP65O2UZ3T/action/author_attestation","sign_citation":"https://pith.science/pith/YFCCBSQE2VLRA7E6OP65O2UZ3T/action/citation_signature","submit_replication":"https://pith.science/pith/YFCCBSQE2VLRA7E6OP65O2UZ3T/action/replication_record"}},"created_at":"2026-07-05T00:10:57.375845+00:00","updated_at":"2026-07-05T00:10:57.375845+00:00"}