{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:LEKKVKKVCK7BXUESMPL6NVQUTX","short_pith_number":"pith:LEKKVKKV","schema_version":"1.0","canonical_sha256":"5914aaa95512be1bd09263d7e6d6149dc287c367cc027943347af8795f4536a4","source":{"kind":"arxiv","id":"2402.02347","version":3},"attestation_state":"computed","paper":{"title":"Riemannian Preconditioned LoRA for Fine-Tuning Foundation Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NA","math.NA","math.OC"],"primary_cat":"cs.LG","authors_text":"Fangzhao Zhang, Mert Pilanci","submitted_at":"2024-02-04T05:05:43Z","abstract_excerpt":"Low-Rank Adaptation (LoRA) emerges as a popular parameter-efficient fine-tuning (PEFT) method, which proposes to freeze pretrained model weights and update an additive low-rank trainable matrix. In this work, we study the enhancement of LoRA training by introducing an $r \\times r$ preconditioner in each gradient step where $r$ is the LoRA rank. We theoretically verify that the proposed preconditioner stabilizes feature learning with LoRA under infinite-width NN setting. Empirically, the implementation of this new preconditioner requires a small change to existing optimizer code and creates vir"},"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":"2402.02347","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-04T05:05:43Z","cross_cats_sorted":["cs.NA","math.NA","math.OC"],"title_canon_sha256":"5b3ca3767a3d766f22e96b3b18b217a646b8133e6a64e68e70ddf340b4ba772a","abstract_canon_sha256":"8bc89ecf7d0dc7ebb3cf9744928296005e3e9070f6a2e173e7fb9cefee768cdb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:27:24.542089Z","signature_b64":"YqT50nhayYwVakwpoB0QcKNb9TL9tTdmOPG4s0pMPQOtAo7rG/brVBjyEa65/TQ6oGsh/+UmBj27asu1lWTDCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5914aaa95512be1bd09263d7e6d6149dc287c367cc027943347af8795f4536a4","last_reissued_at":"2026-07-05T08:27:24.541649Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:27:24.541649Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Riemannian Preconditioned LoRA for Fine-Tuning Foundation Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.NA","math.NA","math.OC"],"primary_cat":"cs.LG","authors_text":"Fangzhao Zhang, Mert Pilanci","submitted_at":"2024-02-04T05:05:43Z","abstract_excerpt":"Low-Rank Adaptation (LoRA) emerges as a popular parameter-efficient fine-tuning (PEFT) method, which proposes to freeze pretrained model weights and update an additive low-rank trainable matrix. In this work, we study the enhancement of LoRA training by introducing an $r \\times r$ preconditioner in each gradient step where $r$ is the LoRA rank. We theoretically verify that the proposed preconditioner stabilizes feature learning with LoRA under infinite-width NN setting. Empirically, the implementation of this new preconditioner requires a small change to existing optimizer code and creates vir"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.02347","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/2402.02347/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":"2402.02347","created_at":"2026-07-05T08:27:24.541702+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.02347v3","created_at":"2026-07-05T08:27:24.541702+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.02347","created_at":"2026-07-05T08:27:24.541702+00:00"},{"alias_kind":"pith_short_12","alias_value":"LEKKVKKVCK7B","created_at":"2026-07-05T08:27:24.541702+00:00"},{"alias_kind":"pith_short_16","alias_value":"LEKKVKKVCK7BXUES","created_at":"2026-07-05T08:27:24.541702+00:00"},{"alias_kind":"pith_short_8","alias_value":"LEKKVKKV","created_at":"2026-07-05T08:27:24.541702+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2602.19926","citing_title":"Rethinking LoRA for Privacy-Preserving Federated Learning in Large Models","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09238","citing_title":"Intrinsic Muon: Spectral Optimization on Riemannian Matrix Manifolds","ref_index":70,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23754","citing_title":"A Retraction-Free EXTRA Method for Decentralized Optimization on the Stiefel Manifold","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LEKKVKKVCK7BXUESMPL6NVQUTX","json":"https://pith.science/pith/LEKKVKKVCK7BXUESMPL6NVQUTX.json","graph_json":"https://pith.science/api/pith-number/LEKKVKKVCK7BXUESMPL6NVQUTX/graph.json","events_json":"https://pith.science/api/pith-number/LEKKVKKVCK7BXUESMPL6NVQUTX/events.json","paper":"https://pith.science/paper/LEKKVKKV"},"agent_actions":{"view_html":"https://pith.science/pith/LEKKVKKVCK7BXUESMPL6NVQUTX","download_json":"https://pith.science/pith/LEKKVKKVCK7BXUESMPL6NVQUTX.json","view_paper":"https://pith.science/paper/LEKKVKKV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.02347&json=true","fetch_graph":"https://pith.science/api/pith-number/LEKKVKKVCK7BXUESMPL6NVQUTX/graph.json","fetch_events":"https://pith.science/api/pith-number/LEKKVKKVCK7BXUESMPL6NVQUTX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LEKKVKKVCK7BXUESMPL6NVQUTX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LEKKVKKVCK7BXUESMPL6NVQUTX/action/storage_attestation","attest_author":"https://pith.science/pith/LEKKVKKVCK7BXUESMPL6NVQUTX/action/author_attestation","sign_citation":"https://pith.science/pith/LEKKVKKVCK7BXUESMPL6NVQUTX/action/citation_signature","submit_replication":"https://pith.science/pith/LEKKVKKVCK7BXUESMPL6NVQUTX/action/replication_record"}},"created_at":"2026-07-05T08:27:24.541702+00:00","updated_at":"2026-07-05T08:27:24.541702+00:00"}