{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:TWANQE5QBCL25A7LHX3R2PSXEW","short_pith_number":"pith:TWANQE5Q","schema_version":"1.0","canonical_sha256":"9d80d813b00897ae83eb3df71d3e5725bba24947769aab8bb3af094d0bac1740","source":{"kind":"arxiv","id":"2204.09560","version":2},"attestation_state":"computed","paper":{"title":"Understanding and Preventing Capacity Loss in Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Clare Lyle, Mark Rowland, Will Dabney","submitted_at":"2022-04-20T15:55:15Z","abstract_excerpt":"The reinforcement learning (RL) problem is rife with sources of non-stationarity, making it a notoriously difficult problem domain for the application of neural networks. We identify a mechanism by which non-stationary prediction targets can prevent learning progress in deep RL agents: \\textit{capacity loss}, whereby networks trained on a sequence of target values lose their ability to quickly update their predictions over time. We demonstrate that capacity loss occurs in a range of RL agents and environments, and is particularly damaging to performance in sparse-reward tasks. We then present "},"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":"2204.09560","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-04-20T15:55:15Z","cross_cats_sorted":[],"title_canon_sha256":"5835991e5ef28cc8429c8777a7d137db325ca656de834380172df9b17eb35fcf","abstract_canon_sha256":"fc0e868588741e16eae7e13936ad2871c4d1e643b4b744a1ad8aec99fdb4060c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:20:19.924790Z","signature_b64":"HFVY4NqUaP5lUMSsDmegbUke1TRX91FKLQ91yRNtp/mnWSfoI9iMb7JFdvINM4sz1sc8eBXtF78dNKpx4NioAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9d80d813b00897ae83eb3df71d3e5725bba24947769aab8bb3af094d0bac1740","last_reissued_at":"2026-07-05T04:20:19.924331Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:20:19.924331Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding and Preventing Capacity Loss in Reinforcement Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Clare Lyle, Mark Rowland, Will Dabney","submitted_at":"2022-04-20T15:55:15Z","abstract_excerpt":"The reinforcement learning (RL) problem is rife with sources of non-stationarity, making it a notoriously difficult problem domain for the application of neural networks. We identify a mechanism by which non-stationary prediction targets can prevent learning progress in deep RL agents: \\textit{capacity loss}, whereby networks trained on a sequence of target values lose their ability to quickly update their predictions over time. We demonstrate that capacity loss occurs in a range of RL agents and environments, and is particularly damaging to performance in sparse-reward tasks. We then present "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2204.09560","kind":"arxiv","version":2},"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/2204.09560/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":"2204.09560","created_at":"2026-07-05T04:20:19.924397+00:00"},{"alias_kind":"arxiv_version","alias_value":"2204.09560v2","created_at":"2026-07-05T04:20:19.924397+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2204.09560","created_at":"2026-07-05T04:20:19.924397+00:00"},{"alias_kind":"pith_short_12","alias_value":"TWANQE5QBCL2","created_at":"2026-07-05T04:20:19.924397+00:00"},{"alias_kind":"pith_short_16","alias_value":"TWANQE5QBCL25A7L","created_at":"2026-07-05T04:20:19.924397+00:00"},{"alias_kind":"pith_short_8","alias_value":"TWANQE5Q","created_at":"2026-07-05T04:20:19.924397+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":8,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07847","citing_title":"When Does Continual Learning Require Learning","ref_index":36,"is_internal_anchor":true},{"citing_arxiv_id":"2606.09762","citing_title":"Preserving Plasticity in Continual Learning via Dynamical Isometry","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2502.20349","citing_title":"Naturalistic Computational Cognitive Science: Towards generalizable models and theories that capture the full range of natural behavior","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2502.20349","citing_title":"Naturalistic Computational Cognitive Science: Towards generalizable models and theories that capture the full range of natural behavior","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2509.22562","citing_title":"Activation Function Design Sustains Plasticity in Continual Learning","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12484","citing_title":"Learning, Fast and Slow: Towards LLMs That Adapt Continually","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12484","citing_title":"Learning, Fast and Slow: Towards LLMs That Adapt Continually","ref_index":35,"is_internal_anchor":false},{"citing_arxiv_id":"2604.15414","citing_title":"Beyond Single-Model Optimization: Preserving Plasticity in Continual Reinforcement Learning","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TWANQE5QBCL25A7LHX3R2PSXEW","json":"https://pith.science/pith/TWANQE5QBCL25A7LHX3R2PSXEW.json","graph_json":"https://pith.science/api/pith-number/TWANQE5QBCL25A7LHX3R2PSXEW/graph.json","events_json":"https://pith.science/api/pith-number/TWANQE5QBCL25A7LHX3R2PSXEW/events.json","paper":"https://pith.science/paper/TWANQE5Q"},"agent_actions":{"view_html":"https://pith.science/pith/TWANQE5QBCL25A7LHX3R2PSXEW","download_json":"https://pith.science/pith/TWANQE5QBCL25A7LHX3R2PSXEW.json","view_paper":"https://pith.science/paper/TWANQE5Q","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2204.09560&json=true","fetch_graph":"https://pith.science/api/pith-number/TWANQE5QBCL25A7LHX3R2PSXEW/graph.json","fetch_events":"https://pith.science/api/pith-number/TWANQE5QBCL25A7LHX3R2PSXEW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TWANQE5QBCL25A7LHX3R2PSXEW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TWANQE5QBCL25A7LHX3R2PSXEW/action/storage_attestation","attest_author":"https://pith.science/pith/TWANQE5QBCL25A7LHX3R2PSXEW/action/author_attestation","sign_citation":"https://pith.science/pith/TWANQE5QBCL25A7LHX3R2PSXEW/action/citation_signature","submit_replication":"https://pith.science/pith/TWANQE5QBCL25A7LHX3R2PSXEW/action/replication_record"}},"created_at":"2026-07-05T04:20:19.924397+00:00","updated_at":"2026-07-05T04:20:19.924397+00:00"}