{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:YUYHDA4MF5ECYOLAKFIKPR436H","short_pith_number":"pith:YUYHDA4M","schema_version":"1.0","canonical_sha256":"c53071838c2f482c39605150a7c79bf1e296a367de4fda9e7353d58b021a9109","source":{"kind":"arxiv","id":"2201.05125","version":3},"attestation_state":"computed","paper":{"title":"GradMax: Growing Neural Networks using Gradient Information","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Bart van Merri\\\"enboer, Fabian Pedregosa, Max Vladymyrov, Thomas Unterthiner, Utku Evci","submitted_at":"2022-01-13T18:30:18Z","abstract_excerpt":"The architecture and the parameters of neural networks are often optimized independently, which requires costly retraining of the parameters whenever the architecture is modified. In this work we instead focus on growing the architecture without requiring costly retraining. We present a method that adds new neurons during training without impacting what is already learned, while improving the training dynamics. We achieve the latter by maximizing the gradients of the new weights and find the optimal initialization efficiently by means of the singular value decomposition (SVD). We call this tec"},"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":"2201.05125","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2022-01-13T18:30:18Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"a91e093de4da8bb6d78c1f50f185944e7a568d7a405fdba19857c083f3443453","abstract_canon_sha256":"e20d3fde55e6617a805aec02ded0089d21dd1c54d36584e4307dd7c24ebf44d3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:29:29.319741Z","signature_b64":"UXa0Olvszaea5EIxQy/HiyKCBQMNOre8D5Jm7p26xFEOhfxWzlyCLIveUTCztgu7nsZaNwWZFqYIhGSQDAXYCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c53071838c2f482c39605150a7c79bf1e296a367de4fda9e7353d58b021a9109","last_reissued_at":"2026-07-05T04:29:29.319226Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:29:29.319226Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GradMax: Growing Neural Networks using Gradient Information","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.LG","authors_text":"Bart van Merri\\\"enboer, Fabian Pedregosa, Max Vladymyrov, Thomas Unterthiner, Utku Evci","submitted_at":"2022-01-13T18:30:18Z","abstract_excerpt":"The architecture and the parameters of neural networks are often optimized independently, which requires costly retraining of the parameters whenever the architecture is modified. In this work we instead focus on growing the architecture without requiring costly retraining. We present a method that adds new neurons during training without impacting what is already learned, while improving the training dynamics. We achieve the latter by maximizing the gradients of the new weights and find the optimal initialization efficiently by means of the singular value decomposition (SVD). We call this tec"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2201.05125","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/2201.05125/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":"2201.05125","created_at":"2026-07-05T04:29:29.319286+00:00"},{"alias_kind":"arxiv_version","alias_value":"2201.05125v3","created_at":"2026-07-05T04:29:29.319286+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2201.05125","created_at":"2026-07-05T04:29:29.319286+00:00"},{"alias_kind":"pith_short_12","alias_value":"YUYHDA4MF5EC","created_at":"2026-07-05T04:29:29.319286+00:00"},{"alias_kind":"pith_short_16","alias_value":"YUYHDA4MF5ECYOLA","created_at":"2026-07-05T04:29:29.319286+00:00"},{"alias_kind":"pith_short_8","alias_value":"YUYHDA4M","created_at":"2026-07-05T04:29:29.319286+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07637","citing_title":"An optimal control approach for neural network architecture adaptation with a posteriori error estimation","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2510.08008","citing_title":"Beyond Sunk Costs: Boosting LLM Pre-training Efficiency via Orthogonal Growth of Mixture-of-Experts","ref_index":4,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YUYHDA4MF5ECYOLAKFIKPR436H","json":"https://pith.science/pith/YUYHDA4MF5ECYOLAKFIKPR436H.json","graph_json":"https://pith.science/api/pith-number/YUYHDA4MF5ECYOLAKFIKPR436H/graph.json","events_json":"https://pith.science/api/pith-number/YUYHDA4MF5ECYOLAKFIKPR436H/events.json","paper":"https://pith.science/paper/YUYHDA4M"},"agent_actions":{"view_html":"https://pith.science/pith/YUYHDA4MF5ECYOLAKFIKPR436H","download_json":"https://pith.science/pith/YUYHDA4MF5ECYOLAKFIKPR436H.json","view_paper":"https://pith.science/paper/YUYHDA4M","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2201.05125&json=true","fetch_graph":"https://pith.science/api/pith-number/YUYHDA4MF5ECYOLAKFIKPR436H/graph.json","fetch_events":"https://pith.science/api/pith-number/YUYHDA4MF5ECYOLAKFIKPR436H/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YUYHDA4MF5ECYOLAKFIKPR436H/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YUYHDA4MF5ECYOLAKFIKPR436H/action/storage_attestation","attest_author":"https://pith.science/pith/YUYHDA4MF5ECYOLAKFIKPR436H/action/author_attestation","sign_citation":"https://pith.science/pith/YUYHDA4MF5ECYOLAKFIKPR436H/action/citation_signature","submit_replication":"https://pith.science/pith/YUYHDA4MF5ECYOLAKFIKPR436H/action/replication_record"}},"created_at":"2026-07-05T04:29:29.319286+00:00","updated_at":"2026-07-05T04:29:29.319286+00:00"}