{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Z7XOP3B47XPWQPI7H3OVTZL2VC","short_pith_number":"pith:Z7XOP3B4","schema_version":"1.0","canonical_sha256":"cfeee7ec3cfddf683d1f3edd59e57aa8b6b881f48a8184b68e6d03420044af34","source":{"kind":"arxiv","id":"2502.01187","version":1},"attestation_state":"computed","paper":{"title":"Skewed Memorization in Large Language Models: Quantification and Decomposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Amir M. Rahmani, Di Huang, Hao Li, Ziyu Wang","submitted_at":"2025-02-03T09:23:53Z","abstract_excerpt":"Memorization in Large Language Models (LLMs) poses privacy and security risks, as models may unintentionally reproduce sensitive or copyrighted data. Existing analyses focus on average-case scenarios, often neglecting the highly skewed distribution of memorization. This paper examines memorization in LLM supervised fine-tuning (SFT), exploring its relationships with training duration, dataset size, and inter-sample similarity. By analyzing memorization probabilities over sequence lengths, we link this skewness to the token generation process, offering insights for estimating memorization and c"},"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":"2502.01187","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2025-02-03T09:23:53Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"5d04d40b1ae33d530c792d17d80f03e741f23a7f745969fd26cca5fe3764580f","abstract_canon_sha256":"c4e104e5459f7f1d6b57e77262e34c14a6a3bdb8a277d76d6a0621da2645c861"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:08:43.935415Z","signature_b64":"dw5qW9YBQL1T8tN5n2wA12Xw4MM1AZYd0ppo63QcRSyHjZSqiORc3XzlADGP7B2x8LMkO7TpmDzVqXe1FpI+AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cfeee7ec3cfddf683d1f3edd59e57aa8b6b881f48a8184b68e6d03420044af34","last_reissued_at":"2026-07-05T10:08:43.935002Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:08:43.935002Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Skewed Memorization in Large Language Models: Quantification and Decomposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.AI","authors_text":"Amir M. Rahmani, Di Huang, Hao Li, Ziyu Wang","submitted_at":"2025-02-03T09:23:53Z","abstract_excerpt":"Memorization in Large Language Models (LLMs) poses privacy and security risks, as models may unintentionally reproduce sensitive or copyrighted data. Existing analyses focus on average-case scenarios, often neglecting the highly skewed distribution of memorization. This paper examines memorization in LLM supervised fine-tuning (SFT), exploring its relationships with training duration, dataset size, and inter-sample similarity. By analyzing memorization probabilities over sequence lengths, we link this skewness to the token generation process, offering insights for estimating memorization and c"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.01187","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/2502.01187/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":"2502.01187","created_at":"2026-07-05T10:08:43.935057+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.01187v1","created_at":"2026-07-05T10:08:43.935057+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.01187","created_at":"2026-07-05T10:08:43.935057+00:00"},{"alias_kind":"pith_short_12","alias_value":"Z7XOP3B47XPW","created_at":"2026-07-05T10:08:43.935057+00:00"},{"alias_kind":"pith_short_16","alias_value":"Z7XOP3B47XPWQPI7","created_at":"2026-07-05T10:08:43.935057+00:00"},{"alias_kind":"pith_short_8","alias_value":"Z7XOP3B4","created_at":"2026-07-05T10:08:43.935057+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17276","citing_title":"On the Memorization Behavior of LLMs in Generative Recommendation: Observations, Implications, and Training Strategies","ref_index":55,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Z7XOP3B47XPWQPI7H3OVTZL2VC","json":"https://pith.science/pith/Z7XOP3B47XPWQPI7H3OVTZL2VC.json","graph_json":"https://pith.science/api/pith-number/Z7XOP3B47XPWQPI7H3OVTZL2VC/graph.json","events_json":"https://pith.science/api/pith-number/Z7XOP3B47XPWQPI7H3OVTZL2VC/events.json","paper":"https://pith.science/paper/Z7XOP3B4"},"agent_actions":{"view_html":"https://pith.science/pith/Z7XOP3B47XPWQPI7H3OVTZL2VC","download_json":"https://pith.science/pith/Z7XOP3B47XPWQPI7H3OVTZL2VC.json","view_paper":"https://pith.science/paper/Z7XOP3B4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.01187&json=true","fetch_graph":"https://pith.science/api/pith-number/Z7XOP3B47XPWQPI7H3OVTZL2VC/graph.json","fetch_events":"https://pith.science/api/pith-number/Z7XOP3B47XPWQPI7H3OVTZL2VC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Z7XOP3B47XPWQPI7H3OVTZL2VC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Z7XOP3B47XPWQPI7H3OVTZL2VC/action/storage_attestation","attest_author":"https://pith.science/pith/Z7XOP3B47XPWQPI7H3OVTZL2VC/action/author_attestation","sign_citation":"https://pith.science/pith/Z7XOP3B47XPWQPI7H3OVTZL2VC/action/citation_signature","submit_replication":"https://pith.science/pith/Z7XOP3B47XPWQPI7H3OVTZL2VC/action/replication_record"}},"created_at":"2026-07-05T10:08:43.935057+00:00","updated_at":"2026-07-05T10:08:43.935057+00:00"}