{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:ESMSOJQY7C2CPLSD76L7IEV7OC","short_pith_number":"pith:ESMSOJQY","schema_version":"1.0","canonical_sha256":"2499272618f8b427ae43ff97f412bf70855ecea634a8b61d6cbf215cef829abd","source":{"kind":"arxiv","id":"2206.07137","version":3},"attestation_state":"computed","paper":{"title":"Prioritized Training on Points that are Learnable, Worth Learning, and Not Yet Learnt","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Adrien Morisot, Aidan N. Gomez, Andreas Kirsch, Benedikt H\\\"oltgen, Jan Brauner, Mrinank Sharma, Muhammed Razzak, Sebastian Farquhar, S\\\"oren Mindermann, Winnie Xu, Yarin Gal","submitted_at":"2022-06-14T19:49:52Z","abstract_excerpt":"Training on web-scale data can take months. But most computation and time is wasted on redundant and noisy points that are already learnt or not learnable. To accelerate training, we introduce Reducible Holdout Loss Selection (RHO-LOSS), a simple but principled technique which selects approximately those points for training that most reduce the model's generalization loss. As a result, RHO-LOSS mitigates the weaknesses of existing data selection methods: techniques from the optimization literature typically select 'hard' (e.g. high loss) points, but such points are often noisy (not learnable) "},"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":"2206.07137","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-06-14T19:49:52Z","cross_cats_sorted":["cs.AI","cs.CL","cs.CV"],"title_canon_sha256":"d3439621fcc21bf1adfc8c7331f1be5bd347c7e5f09631b6792b367c4ef60d08","abstract_canon_sha256":"d9c6201ceabad09a46693fe0ef1be4a6eda0c142cd7508d145f5106449861ad7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:00:43.309674Z","signature_b64":"uMrTMkl9xzqCzWjSiYysLBzZO7ebbhr4nKyJtuFMYejvhxjScBfJUaMFISxCgKfLQRGstvyLIPLBhUvfEimGCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2499272618f8b427ae43ff97f412bf70855ecea634a8b61d6cbf215cef829abd","last_reissued_at":"2026-07-05T05:00:43.309262Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:00:43.309262Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Prioritized Training on Points that are Learnable, Worth Learning, and Not Yet Learnt","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CL","cs.CV"],"primary_cat":"cs.LG","authors_text":"Adrien Morisot, Aidan N. Gomez, Andreas Kirsch, Benedikt H\\\"oltgen, Jan Brauner, Mrinank Sharma, Muhammed Razzak, Sebastian Farquhar, S\\\"oren Mindermann, Winnie Xu, Yarin Gal","submitted_at":"2022-06-14T19:49:52Z","abstract_excerpt":"Training on web-scale data can take months. But most computation and time is wasted on redundant and noisy points that are already learnt or not learnable. To accelerate training, we introduce Reducible Holdout Loss Selection (RHO-LOSS), a simple but principled technique which selects approximately those points for training that most reduce the model's generalization loss. As a result, RHO-LOSS mitigates the weaknesses of existing data selection methods: techniques from the optimization literature typically select 'hard' (e.g. high loss) points, but such points are often noisy (not learnable) "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2206.07137","kind":"arxiv","version":3},"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/2206.07137/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":"2206.07137","created_at":"2026-07-05T05:00:43.309321+00:00"},{"alias_kind":"arxiv_version","alias_value":"2206.07137v3","created_at":"2026-07-05T05:00:43.309321+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2206.07137","created_at":"2026-07-05T05:00:43.309321+00:00"},{"alias_kind":"pith_short_12","alias_value":"ESMSOJQY7C2C","created_at":"2026-07-05T05:00:43.309321+00:00"},{"alias_kind":"pith_short_16","alias_value":"ESMSOJQY7C2CPLSD","created_at":"2026-07-05T05:00:43.309321+00:00"},{"alias_kind":"pith_short_8","alias_value":"ESMSOJQY","created_at":"2026-07-05T05:00:43.309321+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.23969","citing_title":"SLAP: Stratified Loss-based Pruning for On-Policy Data-Efficient Instruction Tuning","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2502.12272","citing_title":"Learning to Reason at the Frontier of Learnability","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2508.04149","citing_title":"Difficulty-Based Preference Data Selection by DPO Implicit Reward Gap","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2509.20786","citing_title":"LiLAW: Lightweight Learnable Adaptive Weighting to Learn Sample Difficulty & Improve Noisy Training","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ESMSOJQY7C2CPLSD76L7IEV7OC","json":"https://pith.science/pith/ESMSOJQY7C2CPLSD76L7IEV7OC.json","graph_json":"https://pith.science/api/pith-number/ESMSOJQY7C2CPLSD76L7IEV7OC/graph.json","events_json":"https://pith.science/api/pith-number/ESMSOJQY7C2CPLSD76L7IEV7OC/events.json","paper":"https://pith.science/paper/ESMSOJQY"},"agent_actions":{"view_html":"https://pith.science/pith/ESMSOJQY7C2CPLSD76L7IEV7OC","download_json":"https://pith.science/pith/ESMSOJQY7C2CPLSD76L7IEV7OC.json","view_paper":"https://pith.science/paper/ESMSOJQY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2206.07137&json=true","fetch_graph":"https://pith.science/api/pith-number/ESMSOJQY7C2CPLSD76L7IEV7OC/graph.json","fetch_events":"https://pith.science/api/pith-number/ESMSOJQY7C2CPLSD76L7IEV7OC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ESMSOJQY7C2CPLSD76L7IEV7OC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ESMSOJQY7C2CPLSD76L7IEV7OC/action/storage_attestation","attest_author":"https://pith.science/pith/ESMSOJQY7C2CPLSD76L7IEV7OC/action/author_attestation","sign_citation":"https://pith.science/pith/ESMSOJQY7C2CPLSD76L7IEV7OC/action/citation_signature","submit_replication":"https://pith.science/pith/ESMSOJQY7C2CPLSD76L7IEV7OC/action/replication_record"}},"created_at":"2026-07-05T05:00:43.309321+00:00","updated_at":"2026-07-05T05:00:43.309321+00:00"}