{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ZEHUMQBGWTAOGNWRKJR6GNVFAL","short_pith_number":"pith:ZEHUMQBG","schema_version":"1.0","canonical_sha256":"c90f464026b4c0e336d15263e336a502dc69f7a13e84493e341c3609b0d2034f","source":{"kind":"arxiv","id":"2503.13751","version":1},"attestation_state":"computed","paper":{"title":"Optimizing ML Training with Metagradient Descent","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Aleksander Madry, Andrew Ilyas, Axel Feldmann, Benjamin Chen, Logan Engstrom, William Moses","submitted_at":"2025-03-17T22:18:24Z","abstract_excerpt":"A major challenge in training large-scale machine learning models is configuring the training process to maximize model performance, i.e., finding the best training setup from a vast design space. In this work, we unlock a gradient-based approach to this problem. We first introduce an algorithm for efficiently calculating metagradients -- gradients through model training -- at scale. We then introduce a \"smooth model training\" framework that enables effective optimization using metagradients. With metagradient descent (MGD), we greatly improve on existing dataset selection methods, outperform "},"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":"2503.13751","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2025-03-17T22:18:24Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"cf531c486802bb534cfbe16b16bc4396b978e62f8d535abd7567913fec3876af","abstract_canon_sha256":"a9365b85a1c053fc4e8a64bab60357a6aa02c93a7be73162f5006c0626021f84"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:33:17.506798Z","signature_b64":"TFOYanPNpt21GOFE/stkLbow3MgCD69KYHYP1/0LerS+qinfRWwrBK62/CHwUr8D3EAiKZ9Byd6xxprzbZ+8DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c90f464026b4c0e336d15263e336a502dc69f7a13e84493e341c3609b0d2034f","last_reissued_at":"2026-07-05T10:33:17.505834Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:33:17.505834Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Optimizing ML Training with Metagradient Descent","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Aleksander Madry, Andrew Ilyas, Axel Feldmann, Benjamin Chen, Logan Engstrom, William Moses","submitted_at":"2025-03-17T22:18:24Z","abstract_excerpt":"A major challenge in training large-scale machine learning models is configuring the training process to maximize model performance, i.e., finding the best training setup from a vast design space. In this work, we unlock a gradient-based approach to this problem. We first introduce an algorithm for efficiently calculating metagradients -- gradients through model training -- at scale. We then introduce a \"smooth model training\" framework that enables effective optimization using metagradients. With metagradient descent (MGD), we greatly improve on existing dataset selection methods, outperform "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.13751","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/2503.13751/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":"2503.13751","created_at":"2026-07-05T10:33:17.505994+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.13751v1","created_at":"2026-07-05T10:33:17.505994+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.13751","created_at":"2026-07-05T10:33:17.505994+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZEHUMQBGWTAO","created_at":"2026-07-05T10:33:17.505994+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZEHUMQBGWTAOGNWR","created_at":"2026-07-05T10:33:17.505994+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZEHUMQBG","created_at":"2026-07-05T10:33:17.505994+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11660","citing_title":"Bergson: An Open Source Library for Data Attribution","ref_index":25,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12572","citing_title":"On the Accuracy of Newton Step and Influence Function Data Attributions","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2602.03783","citing_title":"Efficient Estimation of Kernel Surrogate Models for Task Attribution","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08144","citing_title":"NoiseRater: Meta-Learned Noise Valuation for Diffusion Model Training","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2604.13130","citing_title":"Generalization Guarantees on Data-Driven Tuning of Gradient Descent with Langevin Updates","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2604.07328","citing_title":"How to sketch a learning algorithm","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZEHUMQBGWTAOGNWRKJR6GNVFAL","json":"https://pith.science/pith/ZEHUMQBGWTAOGNWRKJR6GNVFAL.json","graph_json":"https://pith.science/api/pith-number/ZEHUMQBGWTAOGNWRKJR6GNVFAL/graph.json","events_json":"https://pith.science/api/pith-number/ZEHUMQBGWTAOGNWRKJR6GNVFAL/events.json","paper":"https://pith.science/paper/ZEHUMQBG"},"agent_actions":{"view_html":"https://pith.science/pith/ZEHUMQBGWTAOGNWRKJR6GNVFAL","download_json":"https://pith.science/pith/ZEHUMQBGWTAOGNWRKJR6GNVFAL.json","view_paper":"https://pith.science/paper/ZEHUMQBG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.13751&json=true","fetch_graph":"https://pith.science/api/pith-number/ZEHUMQBGWTAOGNWRKJR6GNVFAL/graph.json","fetch_events":"https://pith.science/api/pith-number/ZEHUMQBGWTAOGNWRKJR6GNVFAL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZEHUMQBGWTAOGNWRKJR6GNVFAL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZEHUMQBGWTAOGNWRKJR6GNVFAL/action/storage_attestation","attest_author":"https://pith.science/pith/ZEHUMQBGWTAOGNWRKJR6GNVFAL/action/author_attestation","sign_citation":"https://pith.science/pith/ZEHUMQBGWTAOGNWRKJR6GNVFAL/action/citation_signature","submit_replication":"https://pith.science/pith/ZEHUMQBGWTAOGNWRKJR6GNVFAL/action/replication_record"}},"created_at":"2026-07-05T10:33:17.505994+00:00","updated_at":"2026-07-05T10:33:17.505994+00:00"}