{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:BATXQZ7XQFGLFQ66PE5M52DCUD","short_pith_number":"pith:BATXQZ7X","schema_version":"1.0","canonical_sha256":"08277867f7814cb2c3de793acee862a0debc79a836179a59a46d6a84bf78d732","source":{"kind":"arxiv","id":"2603.04828","version":2},"attestation_state":"computed","paper":{"title":"From Unfamiliar to Familiar: Detecting Pre-training Data via Gradient Deviations in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hainan Zhang, Lingxiang Wang, Ruiqi Zhang, Yanyan Lan, Zhiming Zheng","submitted_at":"2026-03-05T05:21:51Z","abstract_excerpt":"Pre-training data detection for LLMs is essential for addressing copyright concerns and mitigating benchmark contamination. Existing methods mainly focus on the likelihood-based statistical features or heuristic signals before and after fine-tuning, but the former are susceptible to word frequency bias in corpora, and the latter strongly depend on the similarity of fine-tuning data. From an optimization perspective, we observe that during training, samples transition from unfamiliar to familiar in a manner reflected by systematic differences in gradient behavior. Familiar samples exhibit small"},"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":"2603.04828","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-03-05T05:21:51Z","cross_cats_sorted":[],"title_canon_sha256":"28a6c8382b2eae2d58042c64a6a0fe35cb25751f3e6aa9ff250eff15f39d5b04","abstract_canon_sha256":"ad7e636f9c1ab5cd6b4e55a1a43cbfe81b8caf7115979b4070a8e897b23fd27e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T01:03:44.628180Z","signature_b64":"Ssw0B60Ae9gbA9EexlEXWZ01QqKhZDCtrodXKjW6Ivr64Ky6nukul6Jb3mIRfx3d5sqNY2K8dSnHkwzW0gH+BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"08277867f7814cb2c3de793acee862a0debc79a836179a59a46d6a84bf78d732","last_reissued_at":"2026-06-02T01:03:44.627665Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T01:03:44.627665Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From Unfamiliar to Familiar: Detecting Pre-training Data via Gradient Deviations in Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hainan Zhang, Lingxiang Wang, Ruiqi Zhang, Yanyan Lan, Zhiming Zheng","submitted_at":"2026-03-05T05:21:51Z","abstract_excerpt":"Pre-training data detection for LLMs is essential for addressing copyright concerns and mitigating benchmark contamination. Existing methods mainly focus on the likelihood-based statistical features or heuristic signals before and after fine-tuning, but the former are susceptible to word frequency bias in corpora, and the latter strongly depend on the similarity of fine-tuning data. From an optimization perspective, we observe that during training, samples transition from unfamiliar to familiar in a manner reflected by systematic differences in gradient behavior. Familiar samples exhibit small"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.04828","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":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2603.04828/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":"2603.04828","created_at":"2026-06-02T01:03:44.627731+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.04828v2","created_at":"2026-06-02T01:03:44.627731+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.04828","created_at":"2026-06-02T01:03:44.627731+00:00"},{"alias_kind":"pith_short_12","alias_value":"BATXQZ7XQFGL","created_at":"2026-06-02T01:03:44.627731+00:00"},{"alias_kind":"pith_short_16","alias_value":"BATXQZ7XQFGLFQ66","created_at":"2026-06-02T01:03:44.627731+00:00"},{"alias_kind":"pith_short_8","alias_value":"BATXQZ7X","created_at":"2026-06-02T01:03:44.627731+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BATXQZ7XQFGLFQ66PE5M52DCUD","json":"https://pith.science/pith/BATXQZ7XQFGLFQ66PE5M52DCUD.json","graph_json":"https://pith.science/api/pith-number/BATXQZ7XQFGLFQ66PE5M52DCUD/graph.json","events_json":"https://pith.science/api/pith-number/BATXQZ7XQFGLFQ66PE5M52DCUD/events.json","paper":"https://pith.science/paper/BATXQZ7X"},"agent_actions":{"view_html":"https://pith.science/pith/BATXQZ7XQFGLFQ66PE5M52DCUD","download_json":"https://pith.science/pith/BATXQZ7XQFGLFQ66PE5M52DCUD.json","view_paper":"https://pith.science/paper/BATXQZ7X","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.04828&json=true","fetch_graph":"https://pith.science/api/pith-number/BATXQZ7XQFGLFQ66PE5M52DCUD/graph.json","fetch_events":"https://pith.science/api/pith-number/BATXQZ7XQFGLFQ66PE5M52DCUD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BATXQZ7XQFGLFQ66PE5M52DCUD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BATXQZ7XQFGLFQ66PE5M52DCUD/action/storage_attestation","attest_author":"https://pith.science/pith/BATXQZ7XQFGLFQ66PE5M52DCUD/action/author_attestation","sign_citation":"https://pith.science/pith/BATXQZ7XQFGLFQ66PE5M52DCUD/action/citation_signature","submit_replication":"https://pith.science/pith/BATXQZ7XQFGLFQ66PE5M52DCUD/action/replication_record"}},"created_at":"2026-06-02T01:03:44.627731+00:00","updated_at":"2026-06-02T01:03:44.627731+00:00"}