{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:DAAUN7TY746A7XPFGHMV6UXIL3","short_pith_number":"pith:DAAUN7TY","schema_version":"1.0","canonical_sha256":"180146fe78ff3c0fdde531d95f52e85eda5d9d47d0917b84313a45644f1feb21","source":{"kind":"arxiv","id":"2402.15415","version":2},"attestation_state":"computed","paper":{"title":"Understanding Catastrophic Forgetting In LoRA via Mean-Field Attention Dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["math.DS","stat.ML"],"primary_cat":"cs.LG","authors_text":"Hugo Koubbi, Louis Hernandez, Matthieu Boussard","submitted_at":"2024-02-23T16:26:01Z","abstract_excerpt":"Low-Rank Adaptation (LoRA) is the dominant parameter-efficient fine-tuning method due to its favorable compute-performance trade-off, yet it suffers from catastrophic forgetting. We study forgetting through a tractable _mean-field self-attention_ toy model, where tokens evolve as an interacting particle system and LoRA acts as a low-rank perturbation. Using tools from partial differential equations and dynamical systems, we characterize regimes suggesting a phase transition between forgetting and non-forgetting behavior. We show that one phase transition appears with respect to the norm of the"},"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":"2402.15415","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-23T16:26:01Z","cross_cats_sorted":["math.DS","stat.ML"],"title_canon_sha256":"fa31dd8cd30a639e01a70905acc0744e502757cea37f85a88368e92170ce64a5","abstract_canon_sha256":"132cf361e9d93a6f48572a774a5ac763181f414c19a5530fa60bc83d8b46ff86"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-18T02:44:39.763571Z","signature_b64":"fkmr7Xth2gHaXwAhov701BA3tfXD/WY55x5ZM6D0sDsQ5A0e2GPRMQpFAHF5OY2+alfN9E11rwKNXSOvjVRFAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"180146fe78ff3c0fdde531d95f52e85eda5d9d47d0917b84313a45644f1feb21","last_reissued_at":"2026-05-18T02:44:39.762889Z","signature_status":"signed_v1","first_computed_at":"2026-05-18T02:44:39.762889Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding Catastrophic Forgetting In LoRA via Mean-Field Attention Dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["math.DS","stat.ML"],"primary_cat":"cs.LG","authors_text":"Hugo Koubbi, Louis Hernandez, Matthieu Boussard","submitted_at":"2024-02-23T16:26:01Z","abstract_excerpt":"Low-Rank Adaptation (LoRA) is the dominant parameter-efficient fine-tuning method due to its favorable compute-performance trade-off, yet it suffers from catastrophic forgetting. We study forgetting through a tractable _mean-field self-attention_ toy model, where tokens evolve as an interacting particle system and LoRA acts as a low-rank perturbation. Using tools from partial differential equations and dynamical systems, we characterize regimes suggesting a phase transition between forgetting and non-forgetting behavior. We show that one phase transition appears with respect to the norm of the"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.15415","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"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":"2402.15415","created_at":"2026-05-18T02:44:39.762997+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.15415v2","created_at":"2026-05-18T02:44:39.762997+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.15415","created_at":"2026-05-18T02:44:39.762997+00:00"},{"alias_kind":"pith_short_12","alias_value":"DAAUN7TY746A","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"DAAUN7TY746A7XPF","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"DAAUN7TY","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2504.14697","citing_title":"Quantitative Clustering in Mean-Field Transformer Models","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2601.21366","citing_title":"Perceptrons and localization of attention's mean-field landscape","ref_index":11,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DAAUN7TY746A7XPFGHMV6UXIL3","json":"https://pith.science/pith/DAAUN7TY746A7XPFGHMV6UXIL3.json","graph_json":"https://pith.science/api/pith-number/DAAUN7TY746A7XPFGHMV6UXIL3/graph.json","events_json":"https://pith.science/api/pith-number/DAAUN7TY746A7XPFGHMV6UXIL3/events.json","paper":"https://pith.science/paper/DAAUN7TY"},"agent_actions":{"view_html":"https://pith.science/pith/DAAUN7TY746A7XPFGHMV6UXIL3","download_json":"https://pith.science/pith/DAAUN7TY746A7XPFGHMV6UXIL3.json","view_paper":"https://pith.science/paper/DAAUN7TY","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.15415&json=true","fetch_graph":"https://pith.science/api/pith-number/DAAUN7TY746A7XPFGHMV6UXIL3/graph.json","fetch_events":"https://pith.science/api/pith-number/DAAUN7TY746A7XPFGHMV6UXIL3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DAAUN7TY746A7XPFGHMV6UXIL3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DAAUN7TY746A7XPFGHMV6UXIL3/action/storage_attestation","attest_author":"https://pith.science/pith/DAAUN7TY746A7XPFGHMV6UXIL3/action/author_attestation","sign_citation":"https://pith.science/pith/DAAUN7TY746A7XPFGHMV6UXIL3/action/citation_signature","submit_replication":"https://pith.science/pith/DAAUN7TY746A7XPFGHMV6UXIL3/action/replication_record"}},"created_at":"2026-05-18T02:44:39.762997+00:00","updated_at":"2026-05-18T02:44:39.762997+00:00"}