{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:53YMOGFOFU3NLHJC7KIMABOQQG","short_pith_number":"pith:53YMOGFO","schema_version":"1.0","canonical_sha256":"eef0c718ae2d36d59d22fa90c005d08181562ea74fe039e11f6bd94212c1f1c6","source":{"kind":"arxiv","id":"2401.17401","version":1},"attestation_state":"computed","paper":{"title":"Step-size Optimization for Continual Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Arsalan Sharifnassab, Khurram Javed, Richard Sutton, Thomas Degris, Yuxin Liu","submitted_at":"2024-01-30T19:35:43Z","abstract_excerpt":"In continual learning, a learner has to keep learning from the data over its whole life time. A key issue is to decide what knowledge to keep and what knowledge to let go. In a neural network, this can be implemented by using a step-size vector to scale how much gradient samples change network weights. Common algorithms, like RMSProp and Adam, use heuristics, specifically normalization, to adapt this step-size vector. In this paper, we show that those heuristics ignore the effect of their adaptation on the overall objective function, for example by moving the step-size vector away from better "},"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":"2401.17401","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-01-30T19:35:43Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"9a3112ac4d0ef537471e01a7f0afadb5ebf9d85b6782128f8d1acf62a04871e7","abstract_canon_sha256":"95503623a9eb887d07b8703cb285402ac2a1e1266c10c2cd75bf940f2bf5329f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:39:36.247103Z","signature_b64":"xa+KHDc235rk9OEHPDc6HksvXEjEyRFiJqad2OiKgOI0dPaNZOkh+v46h5RLAKZ4m5Brgdh5qIYMXJgK5K7DBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"eef0c718ae2d36d59d22fa90c005d08181562ea74fe039e11f6bd94212c1f1c6","last_reissued_at":"2026-07-05T07:39:36.246619Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:39:36.246619Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Step-size Optimization for Continual Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Arsalan Sharifnassab, Khurram Javed, Richard Sutton, Thomas Degris, Yuxin Liu","submitted_at":"2024-01-30T19:35:43Z","abstract_excerpt":"In continual learning, a learner has to keep learning from the data over its whole life time. A key issue is to decide what knowledge to keep and what knowledge to let go. In a neural network, this can be implemented by using a step-size vector to scale how much gradient samples change network weights. Common algorithms, like RMSProp and Adam, use heuristics, specifically normalization, to adapt this step-size vector. In this paper, we show that those heuristics ignore the effect of their adaptation on the overall objective function, for example by moving the step-size vector away from better "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.17401","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/2401.17401/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":"2401.17401","created_at":"2026-07-05T07:39:36.246701+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.17401v1","created_at":"2026-07-05T07:39:36.246701+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.17401","created_at":"2026-07-05T07:39:36.246701+00:00"},{"alias_kind":"pith_short_12","alias_value":"53YMOGFOFU3N","created_at":"2026-07-05T07:39:36.246701+00:00"},{"alias_kind":"pith_short_16","alias_value":"53YMOGFOFU3NLHJC","created_at":"2026-07-05T07:39:36.246701+00:00"},{"alias_kind":"pith_short_8","alias_value":"53YMOGFO","created_at":"2026-07-05T07:39:36.246701+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.04029","citing_title":"Position: Deployed Reinforcement Learning should be Continual","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.23061","citing_title":"Anytime Training with Schedule-Free Spectral Optimization","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.27063","citing_title":"Learning to Forget: Continual Learning with Adaptive Weight Decay","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06764","citing_title":"Revisiting Adam for Streaming Reinforcement Learning","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/53YMOGFOFU3NLHJC7KIMABOQQG","json":"https://pith.science/pith/53YMOGFOFU3NLHJC7KIMABOQQG.json","graph_json":"https://pith.science/api/pith-number/53YMOGFOFU3NLHJC7KIMABOQQG/graph.json","events_json":"https://pith.science/api/pith-number/53YMOGFOFU3NLHJC7KIMABOQQG/events.json","paper":"https://pith.science/paper/53YMOGFO"},"agent_actions":{"view_html":"https://pith.science/pith/53YMOGFOFU3NLHJC7KIMABOQQG","download_json":"https://pith.science/pith/53YMOGFOFU3NLHJC7KIMABOQQG.json","view_paper":"https://pith.science/paper/53YMOGFO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.17401&json=true","fetch_graph":"https://pith.science/api/pith-number/53YMOGFOFU3NLHJC7KIMABOQQG/graph.json","fetch_events":"https://pith.science/api/pith-number/53YMOGFOFU3NLHJC7KIMABOQQG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/53YMOGFOFU3NLHJC7KIMABOQQG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/53YMOGFOFU3NLHJC7KIMABOQQG/action/storage_attestation","attest_author":"https://pith.science/pith/53YMOGFOFU3NLHJC7KIMABOQQG/action/author_attestation","sign_citation":"https://pith.science/pith/53YMOGFOFU3NLHJC7KIMABOQQG/action/citation_signature","submit_replication":"https://pith.science/pith/53YMOGFOFU3NLHJC7KIMABOQQG/action/replication_record"}},"created_at":"2026-07-05T07:39:36.246701+00:00","updated_at":"2026-07-05T07:39:36.246701+00:00"}