{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:M7IJ3EYNKX4LQOPSTMUHEZ5XF2","short_pith_number":"pith:M7IJ3EYN","schema_version":"1.0","canonical_sha256":"67d09d930d55f8b839f29b287267b72e81fedf3978b8d060b402f6943e638e0a","source":{"kind":"arxiv","id":"2207.00986","version":1},"attestation_state":"computed","paper":{"title":"Stabilizing Off-Policy Deep Reinforcement Learning from Pixels","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Edoardo Cetin, Oya Celiktutan, Philip J. Ball, Steve Roberts","submitted_at":"2022-07-03T08:52:40Z","abstract_excerpt":"Off-policy reinforcement learning (RL) from pixel observations is notoriously unstable. As a result, many successful algorithms must combine different domain-specific practices and auxiliary losses to learn meaningful behaviors in complex environments. In this work, we provide novel analysis demonstrating that these instabilities arise from performing temporal-difference learning with a convolutional encoder and low-magnitude rewards. We show that this new visual deadly triad causes unstable training and premature convergence to degenerate solutions, a phenomenon we name catastrophic self-over"},"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":"2207.00986","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2022-07-03T08:52:40Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"749f4748816750180fe1ad557d3f06ed381bceb4b4adb82ebd7512d6af4f55ce","abstract_canon_sha256":"49a9b7545214df70a346264bf05d13757cadb9b549ae89adad44439ffd63735a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:36:54.806569Z","signature_b64":"RkQEGT8rG5w/C3LNIQatLO5qO2Gn5OWI/Wn8utUO3ey/umTNciVhe0Oj7FqZPY5HauRfGwIea/M8UXHW7suXAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"67d09d930d55f8b839f29b287267b72e81fedf3978b8d060b402f6943e638e0a","last_reissued_at":"2026-07-05T04:36:54.806148Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:36:54.806148Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stabilizing Off-Policy Deep Reinforcement Learning from Pixels","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Edoardo Cetin, Oya Celiktutan, Philip J. Ball, Steve Roberts","submitted_at":"2022-07-03T08:52:40Z","abstract_excerpt":"Off-policy reinforcement learning (RL) from pixel observations is notoriously unstable. As a result, many successful algorithms must combine different domain-specific practices and auxiliary losses to learn meaningful behaviors in complex environments. In this work, we provide novel analysis demonstrating that these instabilities arise from performing temporal-difference learning with a convolutional encoder and low-magnitude rewards. We show that this new visual deadly triad causes unstable training and premature convergence to degenerate solutions, a phenomenon we name catastrophic self-over"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2207.00986","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/2207.00986/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":"2207.00986","created_at":"2026-07-05T04:36:54.806203+00:00"},{"alias_kind":"arxiv_version","alias_value":"2207.00986v1","created_at":"2026-07-05T04:36:54.806203+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2207.00986","created_at":"2026-07-05T04:36:54.806203+00:00"},{"alias_kind":"pith_short_12","alias_value":"M7IJ3EYNKX4L","created_at":"2026-07-05T04:36:54.806203+00:00"},{"alias_kind":"pith_short_16","alias_value":"M7IJ3EYNKX4LQOPS","created_at":"2026-07-05T04:36:54.806203+00:00"},{"alias_kind":"pith_short_8","alias_value":"M7IJ3EYN","created_at":"2026-07-05T04:36:54.806203+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2310.02635","citing_title":"Reinforcement Learning with Foundation Priors: Let the Embodied Agent Efficiently Learn on Its Own","ref_index":70,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/M7IJ3EYNKX4LQOPSTMUHEZ5XF2","json":"https://pith.science/pith/M7IJ3EYNKX4LQOPSTMUHEZ5XF2.json","graph_json":"https://pith.science/api/pith-number/M7IJ3EYNKX4LQOPSTMUHEZ5XF2/graph.json","events_json":"https://pith.science/api/pith-number/M7IJ3EYNKX4LQOPSTMUHEZ5XF2/events.json","paper":"https://pith.science/paper/M7IJ3EYN"},"agent_actions":{"view_html":"https://pith.science/pith/M7IJ3EYNKX4LQOPSTMUHEZ5XF2","download_json":"https://pith.science/pith/M7IJ3EYNKX4LQOPSTMUHEZ5XF2.json","view_paper":"https://pith.science/paper/M7IJ3EYN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2207.00986&json=true","fetch_graph":"https://pith.science/api/pith-number/M7IJ3EYNKX4LQOPSTMUHEZ5XF2/graph.json","fetch_events":"https://pith.science/api/pith-number/M7IJ3EYNKX4LQOPSTMUHEZ5XF2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/M7IJ3EYNKX4LQOPSTMUHEZ5XF2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/M7IJ3EYNKX4LQOPSTMUHEZ5XF2/action/storage_attestation","attest_author":"https://pith.science/pith/M7IJ3EYNKX4LQOPSTMUHEZ5XF2/action/author_attestation","sign_citation":"https://pith.science/pith/M7IJ3EYNKX4LQOPSTMUHEZ5XF2/action/citation_signature","submit_replication":"https://pith.science/pith/M7IJ3EYNKX4LQOPSTMUHEZ5XF2/action/replication_record"}},"created_at":"2026-07-05T04:36:54.806203+00:00","updated_at":"2026-07-05T04:36:54.806203+00:00"}