{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:HXXOABWOYCPRQURM6SUGQIJUBS","short_pith_number":"pith:HXXOABWO","schema_version":"1.0","canonical_sha256":"3deee006cec09f18522cf4a86821340c984f1d8a155e9e459c8c488a7f87938e","source":{"kind":"arxiv","id":"2205.07802","version":1},"attestation_state":"computed","paper":{"title":"The Primacy Bias in Deep Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Aaron Courville, Evgenii Nikishin, Max Schwarzer, Pierluca D'Oro, Pierre-Luc Bacon","submitted_at":"2022-05-16T16:48:36Z","abstract_excerpt":"This work identifies a common flaw of deep reinforcement learning (RL) algorithms: a tendency to rely on early interactions and ignore useful evidence encountered later. Because of training on progressively growing datasets, deep RL agents incur a risk of overfitting to earlier experiences, negatively affecting the rest of the learning process. Inspired by cognitive science, we refer to this effect as the primacy bias. Through a series of experiments, we dissect the algorithmic aspects of deep RL that exacerbate this bias. We then propose a simple yet generally-applicable mechanism that tackle"},"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":"2205.07802","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2022-05-16T16:48:36Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"e308fc9c924bace9015b71990b7979b1882b500c9ae87588e9a25f8d541d7e16","abstract_canon_sha256":"725723158f3504d8644e2c933cbad76a63159c39dc7b401fdcc421099ba68e80"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:23:32.331343Z","signature_b64":"weq51jNEVzJEAUW/SUFJNJggfoJ4zKDXe/GzWsgX3EnpS10DP6J2Uloy10rK0ZeYSTQga+/QGpg3QFkAHG/hAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3deee006cec09f18522cf4a86821340c984f1d8a155e9e459c8c488a7f87938e","last_reissued_at":"2026-07-05T04:23:32.330961Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:23:32.330961Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Primacy Bias in Deep Reinforcement Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Aaron Courville, Evgenii Nikishin, Max Schwarzer, Pierluca D'Oro, Pierre-Luc Bacon","submitted_at":"2022-05-16T16:48:36Z","abstract_excerpt":"This work identifies a common flaw of deep reinforcement learning (RL) algorithms: a tendency to rely on early interactions and ignore useful evidence encountered later. Because of training on progressively growing datasets, deep RL agents incur a risk of overfitting to earlier experiences, negatively affecting the rest of the learning process. Inspired by cognitive science, we refer to this effect as the primacy bias. Through a series of experiments, we dissect the algorithmic aspects of deep RL that exacerbate this bias. We then propose a simple yet generally-applicable mechanism that tackle"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.07802","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/2205.07802/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":"2205.07802","created_at":"2026-07-05T04:23:32.331018+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.07802v1","created_at":"2026-07-05T04:23:32.331018+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.07802","created_at":"2026-07-05T04:23:32.331018+00:00"},{"alias_kind":"pith_short_12","alias_value":"HXXOABWOYCPR","created_at":"2026-07-05T04:23:32.331018+00:00"},{"alias_kind":"pith_short_16","alias_value":"HXXOABWOYCPRQURM","created_at":"2026-07-05T04:23:32.331018+00:00"},{"alias_kind":"pith_short_8","alias_value":"HXXOABWO","created_at":"2026-07-05T04:23:32.331018+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18487","citing_title":"SFT Overtraining Predicts Rank Inversion via Entropy Collapse Under RLVR","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23565","citing_title":"Understanding Goal Generalisation in Sequential Reinforcement Learning","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09028","citing_title":"Plasticity-Enhanced Multi-Agent Mixture of Experts for Dynamic Objective Adaptation in UAVs-Assisted Emergency Communication Networks","ref_index":12,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HXXOABWOYCPRQURM6SUGQIJUBS","json":"https://pith.science/pith/HXXOABWOYCPRQURM6SUGQIJUBS.json","graph_json":"https://pith.science/api/pith-number/HXXOABWOYCPRQURM6SUGQIJUBS/graph.json","events_json":"https://pith.science/api/pith-number/HXXOABWOYCPRQURM6SUGQIJUBS/events.json","paper":"https://pith.science/paper/HXXOABWO"},"agent_actions":{"view_html":"https://pith.science/pith/HXXOABWOYCPRQURM6SUGQIJUBS","download_json":"https://pith.science/pith/HXXOABWOYCPRQURM6SUGQIJUBS.json","view_paper":"https://pith.science/paper/HXXOABWO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.07802&json=true","fetch_graph":"https://pith.science/api/pith-number/HXXOABWOYCPRQURM6SUGQIJUBS/graph.json","fetch_events":"https://pith.science/api/pith-number/HXXOABWOYCPRQURM6SUGQIJUBS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HXXOABWOYCPRQURM6SUGQIJUBS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HXXOABWOYCPRQURM6SUGQIJUBS/action/storage_attestation","attest_author":"https://pith.science/pith/HXXOABWOYCPRQURM6SUGQIJUBS/action/author_attestation","sign_citation":"https://pith.science/pith/HXXOABWOYCPRQURM6SUGQIJUBS/action/citation_signature","submit_replication":"https://pith.science/pith/HXXOABWOYCPRQURM6SUGQIJUBS/action/replication_record"}},"created_at":"2026-07-05T04:23:32.331018+00:00","updated_at":"2026-07-05T04:23:32.331018+00:00"}