{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:4NWABOF5UHTJQOMVQNN6RVVWB3","short_pith_number":"pith:4NWABOF5","schema_version":"1.0","canonical_sha256":"e36c00b8bda1e6983995835be8d6b60ed9937553f8ea205d628c42621c67d0cf","source":{"kind":"arxiv","id":"1903.08894","version":1},"attestation_state":"computed","paper":{"title":"Towards Characterizing Divergence in Deep Q-Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ethan Knight, Joshua Achiam, Pieter Abbeel","submitted_at":"2019-03-21T09:42:41Z","abstract_excerpt":"Deep Q-Learning (DQL), a family of temporal difference algorithms for control, employs three techniques collectively known as the `deadly triad' in reinforcement learning: bootstrapping, off-policy learning, and function approximation. Prior work has demonstrated that together these can lead to divergence in Q-learning algorithms, but the conditions under which divergence occurs are not well-understood. In this note, we give a simple analysis based on a linear approximation to the Q-value updates, which we believe provides insight into divergence under the deadly triad. The central point in ou"},"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":"1903.08894","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2019-03-21T09:42:41Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2cb9ccea37a2aeda9425c2ce2279ab7d1999ef891a4550d7b54b8d250ce8b675","abstract_canon_sha256":"893a883be8f7977e05faffce176fd26bb6af2d744dcd6397e74544ecebee37ee"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:50:43.647121Z","signature_b64":"ETGLZRyip5iyyP9XU2ZF0vHJ3dA15AbvdlEEqzTPpdX1MHpHuRzQpnHf+gAFNdZnJVa6SbY0+2go7imgW75IDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e36c00b8bda1e6983995835be8d6b60ed9937553f8ea205d628c42621c67d0cf","last_reissued_at":"2026-05-17T23:50:43.646611Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:50:43.646611Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Characterizing Divergence in Deep Q-Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ethan Knight, Joshua Achiam, Pieter Abbeel","submitted_at":"2019-03-21T09:42:41Z","abstract_excerpt":"Deep Q-Learning (DQL), a family of temporal difference algorithms for control, employs three techniques collectively known as the `deadly triad' in reinforcement learning: bootstrapping, off-policy learning, and function approximation. Prior work has demonstrated that together these can lead to divergence in Q-learning algorithms, but the conditions under which divergence occurs are not well-understood. In this note, we give a simple analysis based on a linear approximation to the Q-value updates, which we believe provides insight into divergence under the deadly triad. The central point in ou"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1903.08894","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":""},"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":"1903.08894","created_at":"2026-05-17T23:50:43.646693+00:00"},{"alias_kind":"arxiv_version","alias_value":"1903.08894v1","created_at":"2026-05-17T23:50:43.646693+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1903.08894","created_at":"2026-05-17T23:50:43.646693+00:00"},{"alias_kind":"pith_short_12","alias_value":"4NWABOF5UHTJ","created_at":"2026-05-18T12:33:10.108867+00:00"},{"alias_kind":"pith_short_16","alias_value":"4NWABOF5UHTJQOMV","created_at":"2026-05-18T12:33:10.108867+00:00"},{"alias_kind":"pith_short_8","alias_value":"4NWABOF5","created_at":"2026-05-18T12:33:10.108867+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.01968","citing_title":"AdamO: A Collapse-Suppressed Optimizer for Offline RL","ref_index":39,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4NWABOF5UHTJQOMVQNN6RVVWB3","json":"https://pith.science/pith/4NWABOF5UHTJQOMVQNN6RVVWB3.json","graph_json":"https://pith.science/api/pith-number/4NWABOF5UHTJQOMVQNN6RVVWB3/graph.json","events_json":"https://pith.science/api/pith-number/4NWABOF5UHTJQOMVQNN6RVVWB3/events.json","paper":"https://pith.science/paper/4NWABOF5"},"agent_actions":{"view_html":"https://pith.science/pith/4NWABOF5UHTJQOMVQNN6RVVWB3","download_json":"https://pith.science/pith/4NWABOF5UHTJQOMVQNN6RVVWB3.json","view_paper":"https://pith.science/paper/4NWABOF5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1903.08894&json=true","fetch_graph":"https://pith.science/api/pith-number/4NWABOF5UHTJQOMVQNN6RVVWB3/graph.json","fetch_events":"https://pith.science/api/pith-number/4NWABOF5UHTJQOMVQNN6RVVWB3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4NWABOF5UHTJQOMVQNN6RVVWB3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4NWABOF5UHTJQOMVQNN6RVVWB3/action/storage_attestation","attest_author":"https://pith.science/pith/4NWABOF5UHTJQOMVQNN6RVVWB3/action/author_attestation","sign_citation":"https://pith.science/pith/4NWABOF5UHTJQOMVQNN6RVVWB3/action/citation_signature","submit_replication":"https://pith.science/pith/4NWABOF5UHTJQOMVQNN6RVVWB3/action/replication_record"}},"created_at":"2026-05-17T23:50:43.646693+00:00","updated_at":"2026-05-17T23:50:43.646693+00:00"}