{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:H5HSXINO4QZJETQQV5G7PZODHZ","short_pith_number":"pith:H5HSXINO","schema_version":"1.0","canonical_sha256":"3f4f2ba1aee432924e10af4df7e5c33e7628dbca56646363991872574a2fa340","source":{"kind":"arxiv","id":"2411.15385","version":1},"attestation_state":"computed","paper":{"title":"Gradient dynamics for low-rank fine-tuning beyond kernels","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.ST","stat.ML","stat.TH"],"primary_cat":"cs.LG","authors_text":"Arif Kerem Dayi, Sitan Chen","submitted_at":"2024-11-23T00:00:28Z","abstract_excerpt":"LoRA has emerged as one of the de facto methods for fine-tuning foundation models with low computational cost and memory footprint. The idea is to only train a low-rank perturbation to the weights of a pre-trained model, given supervised data for a downstream task. Despite its empirical sucess, from a mathematical perspective it remains poorly understood what learning mechanisms ensure that gradient descent converges to useful low-rank perturbations.\n  In this work we study low-rank fine-tuning in a student-teacher setting. We are given the weights of a two-layer base model $f$, as well as i.i"},"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":"2411.15385","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-11-23T00:00:28Z","cross_cats_sorted":["math.ST","stat.ML","stat.TH"],"title_canon_sha256":"e050a4adadfb2713014f623206381d128f8bfdd9a879a001aea010bc9b6f34d9","abstract_canon_sha256":"51f833e8c4b8c17adc166935129057ce9da583c07794c4c10dcb2b41cde01c53"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:39:33.865904Z","signature_b64":"8aRkUNe3Bs0LD34uJfNvP9iuKvbq+psAqDb2rIWYrt22X5l0QST+9fQfKvgcLV0yEr1aF12jm8ALBhv0G/sfBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3f4f2ba1aee432924e10af4df7e5c33e7628dbca56646363991872574a2fa340","last_reissued_at":"2026-07-05T09:39:33.865490Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:39:33.865490Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Gradient dynamics for low-rank fine-tuning beyond kernels","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.ST","stat.ML","stat.TH"],"primary_cat":"cs.LG","authors_text":"Arif Kerem Dayi, Sitan Chen","submitted_at":"2024-11-23T00:00:28Z","abstract_excerpt":"LoRA has emerged as one of the de facto methods for fine-tuning foundation models with low computational cost and memory footprint. The idea is to only train a low-rank perturbation to the weights of a pre-trained model, given supervised data for a downstream task. Despite its empirical sucess, from a mathematical perspective it remains poorly understood what learning mechanisms ensure that gradient descent converges to useful low-rank perturbations.\n  In this work we study low-rank fine-tuning in a student-teacher setting. We are given the weights of a two-layer base model $f$, as well as i.i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.15385","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/2411.15385/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":"2411.15385","created_at":"2026-07-05T09:39:33.865548+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.15385v1","created_at":"2026-07-05T09:39:33.865548+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.15385","created_at":"2026-07-05T09:39:33.865548+00:00"},{"alias_kind":"pith_short_12","alias_value":"H5HSXINO4QZJ","created_at":"2026-07-05T09:39:33.865548+00:00"},{"alias_kind":"pith_short_16","alias_value":"H5HSXINO4QZJETQQ","created_at":"2026-07-05T09:39:33.865548+00:00"},{"alias_kind":"pith_short_8","alias_value":"H5HSXINO","created_at":"2026-07-05T09:39:33.865548+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.01656","citing_title":"Mixture of Experts Provably Detect and Learn the Latent Cluster Structure in Gradient-Based Learning","ref_index":1262,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/H5HSXINO4QZJETQQV5G7PZODHZ","json":"https://pith.science/pith/H5HSXINO4QZJETQQV5G7PZODHZ.json","graph_json":"https://pith.science/api/pith-number/H5HSXINO4QZJETQQV5G7PZODHZ/graph.json","events_json":"https://pith.science/api/pith-number/H5HSXINO4QZJETQQV5G7PZODHZ/events.json","paper":"https://pith.science/paper/H5HSXINO"},"agent_actions":{"view_html":"https://pith.science/pith/H5HSXINO4QZJETQQV5G7PZODHZ","download_json":"https://pith.science/pith/H5HSXINO4QZJETQQV5G7PZODHZ.json","view_paper":"https://pith.science/paper/H5HSXINO","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.15385&json=true","fetch_graph":"https://pith.science/api/pith-number/H5HSXINO4QZJETQQV5G7PZODHZ/graph.json","fetch_events":"https://pith.science/api/pith-number/H5HSXINO4QZJETQQV5G7PZODHZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/H5HSXINO4QZJETQQV5G7PZODHZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/H5HSXINO4QZJETQQV5G7PZODHZ/action/storage_attestation","attest_author":"https://pith.science/pith/H5HSXINO4QZJETQQV5G7PZODHZ/action/author_attestation","sign_citation":"https://pith.science/pith/H5HSXINO4QZJETQQV5G7PZODHZ/action/citation_signature","submit_replication":"https://pith.science/pith/H5HSXINO4QZJETQQV5G7PZODHZ/action/replication_record"}},"created_at":"2026-07-05T09:39:33.865548+00:00","updated_at":"2026-07-05T09:39:33.865548+00:00"}