{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:2WKYZAIP32E7YTURREN4MZO2BH","short_pith_number":"pith:2WKYZAIP","schema_version":"1.0","canonical_sha256":"d5958c810fde89fc4e91891bc665da09e633b29513d6a6d73c55c5565788e4a9","source":{"kind":"arxiv","id":"1912.07942","version":4},"attestation_state":"computed","paper":{"title":"Analyzing Information Leakage of Updates to Natural Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CR","stat.ML"],"primary_cat":"cs.LG","authors_text":"Andrew Paverd, Boris K\\\"opf, Lukas Wutschitz, Marc Brockschmidt, Olga Ohrimenko, Santiago Zanella-B\\'eguelin, Shruti Tople, Victor R\\\"uhle","submitted_at":"2019-12-17T11:46:08Z","abstract_excerpt":"To continuously improve quality and reflect changes in data, machine learning applications have to regularly retrain and update their core models. We show that a differential analysis of language model snapshots before and after an update can reveal a surprising amount of detailed information about changes in the training data. We propose two new metrics---\\emph{differential score} and \\emph{differential rank}---for analyzing the leakage due to updates of natural language models. We perform leakage analysis using these metrics across models trained on several different datasets using different"},"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":"1912.07942","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2019-12-17T11:46:08Z","cross_cats_sorted":["cs.CL","cs.CR","stat.ML"],"title_canon_sha256":"b0b0ee56dd1cf78369ccd10fe8aa8862ebfa62e1f7a93082b577c0e41fd22ad2","abstract_canon_sha256":"3ed887bc633bf97cc636e8f12608fa86eb1cb6fc974f0079686040c42340d5e5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:03:37.438135Z","signature_b64":"5OAj3Ve5sLbeYmIL/vQjZVresQvTNGuC2xNAbusgVdLNaOcvQBtQ65a0s0gh3K8kE5GjuVEroXx1sPBS66mqDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d5958c810fde89fc4e91891bc665da09e633b29513d6a6d73c55c5565788e4a9","last_reissued_at":"2026-07-05T03:03:37.437569Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:03:37.437569Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Analyzing Information Leakage of Updates to Natural Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.CR","stat.ML"],"primary_cat":"cs.LG","authors_text":"Andrew Paverd, Boris K\\\"opf, Lukas Wutschitz, Marc Brockschmidt, Olga Ohrimenko, Santiago Zanella-B\\'eguelin, Shruti Tople, Victor R\\\"uhle","submitted_at":"2019-12-17T11:46:08Z","abstract_excerpt":"To continuously improve quality and reflect changes in data, machine learning applications have to regularly retrain and update their core models. We show that a differential analysis of language model snapshots before and after an update can reveal a surprising amount of detailed information about changes in the training data. We propose two new metrics---\\emph{differential score} and \\emph{differential rank}---for analyzing the leakage due to updates of natural language models. We perform leakage analysis using these metrics across models trained on several different datasets using different"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1912.07942","kind":"arxiv","version":4},"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/1912.07942/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":"1912.07942","created_at":"2026-07-05T03:03:37.437633+00:00"},{"alias_kind":"arxiv_version","alias_value":"1912.07942v4","created_at":"2026-07-05T03:03:37.437633+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1912.07942","created_at":"2026-07-05T03:03:37.437633+00:00"},{"alias_kind":"pith_short_12","alias_value":"2WKYZAIP32E7","created_at":"2026-07-05T03:03:37.437633+00:00"},{"alias_kind":"pith_short_16","alias_value":"2WKYZAIP32E7YTUR","created_at":"2026-07-05T03:03:37.437633+00:00"},{"alias_kind":"pith_short_8","alias_value":"2WKYZAIP","created_at":"2026-07-05T03:03:37.437633+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2502.09288","citing_title":"AI Safety for Everyone","ref_index":89,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2WKYZAIP32E7YTURREN4MZO2BH","json":"https://pith.science/pith/2WKYZAIP32E7YTURREN4MZO2BH.json","graph_json":"https://pith.science/api/pith-number/2WKYZAIP32E7YTURREN4MZO2BH/graph.json","events_json":"https://pith.science/api/pith-number/2WKYZAIP32E7YTURREN4MZO2BH/events.json","paper":"https://pith.science/paper/2WKYZAIP"},"agent_actions":{"view_html":"https://pith.science/pith/2WKYZAIP32E7YTURREN4MZO2BH","download_json":"https://pith.science/pith/2WKYZAIP32E7YTURREN4MZO2BH.json","view_paper":"https://pith.science/paper/2WKYZAIP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1912.07942&json=true","fetch_graph":"https://pith.science/api/pith-number/2WKYZAIP32E7YTURREN4MZO2BH/graph.json","fetch_events":"https://pith.science/api/pith-number/2WKYZAIP32E7YTURREN4MZO2BH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2WKYZAIP32E7YTURREN4MZO2BH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2WKYZAIP32E7YTURREN4MZO2BH/action/storage_attestation","attest_author":"https://pith.science/pith/2WKYZAIP32E7YTURREN4MZO2BH/action/author_attestation","sign_citation":"https://pith.science/pith/2WKYZAIP32E7YTURREN4MZO2BH/action/citation_signature","submit_replication":"https://pith.science/pith/2WKYZAIP32E7YTURREN4MZO2BH/action/replication_record"}},"created_at":"2026-07-05T03:03:37.437633+00:00","updated_at":"2026-07-05T03:03:37.437633+00:00"}