{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:JZ7Y2DXFLTJZOYEAYCTTQVOUGH","short_pith_number":"pith:JZ7Y2DXF","schema_version":"1.0","canonical_sha256":"4e7f8d0ee55cd3976080c0a73855d431fe92aa139eca243c6cd4f3a49ad035d3","source":{"kind":"arxiv","id":"2607.14205","version":1},"attestation_state":"computed","paper":{"title":"Privacy Leakage in Federated Learning in Radiology Reports: A Comparative Evaluation of Tokenizer-Driven Privacy Risks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Andres Martinez, Dimitrios Bounias, Klaus Maier-Hein, Ralf Floca, Santhosh Parampottupadam, Sinem Sav","submitted_at":"2026-07-15T17:57:53Z","abstract_excerpt":"Federated learning (FL) enables multi-institutional training on clinical text without sharing raw data, but gradient inversion can reconstruct sensitive information from shared model updates. The extent of this leakage for radiology reports, and the role of tokenizer design, remains unclear. We quantify gradient-based text reconstruction in FL and compare privacy risk across three tokenizers with the model architecture held fixed. Six FL clients trained a GPT-2-style transformer (sequence length 32) on public radiology corpora (368,751 diagnostic reports, 98,206 discharge summaries, 1,500 MIMI"},"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":"2607.14205","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-07-15T17:57:53Z","cross_cats_sorted":["cs.CL","cs.CR"],"title_canon_sha256":"53b353e96036b673a01ac41e94d61e668ef498fe760eb76130e13c4fb4b85509","abstract_canon_sha256":"2ecef1f772ce5929134227dadaf2245b8b1bc0ca2c3bbbf2c9a551249c2192f4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-17T00:20:57.978349Z","signature_b64":"za3v2Vbywa9t6X57+rw2IVMow8d1TqhIAZFzGTkVEmxIC92gUgyKMr5zuDciYP+dcQMF9fWUZPqXwvTKu0XlAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4e7f8d0ee55cd3976080c0a73855d431fe92aa139eca243c6cd4f3a49ad035d3","last_reissued_at":"2026-07-17T00:20:57.977522Z","signature_status":"signed_v1","first_computed_at":"2026-07-17T00:20:57.977522Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Privacy Leakage in Federated Learning in Radiology Reports: A Comparative Evaluation of Tokenizer-Driven Privacy Risks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL","cs.CR"],"primary_cat":"cs.LG","authors_text":"Andres Martinez, Dimitrios Bounias, Klaus Maier-Hein, Ralf Floca, Santhosh Parampottupadam, Sinem Sav","submitted_at":"2026-07-15T17:57:53Z","abstract_excerpt":"Federated learning (FL) enables multi-institutional training on clinical text without sharing raw data, but gradient inversion can reconstruct sensitive information from shared model updates. The extent of this leakage for radiology reports, and the role of tokenizer design, remains unclear. We quantify gradient-based text reconstruction in FL and compare privacy risk across three tokenizers with the model architecture held fixed. Six FL clients trained a GPT-2-style transformer (sequence length 32) on public radiology corpora (368,751 diagnostic reports, 98,206 discharge summaries, 1,500 MIMI"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.14205","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/2607.14205/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":"2607.14205","created_at":"2026-07-17T00:20:57.977932+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.14205v1","created_at":"2026-07-17T00:20:57.977932+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.14205","created_at":"2026-07-17T00:20:57.977932+00:00"},{"alias_kind":"pith_short_12","alias_value":"JZ7Y2DXFLTJZ","created_at":"2026-07-17T00:20:57.977932+00:00"},{"alias_kind":"pith_short_16","alias_value":"JZ7Y2DXFLTJZOYEA","created_at":"2026-07-17T00:20:57.977932+00:00"},{"alias_kind":"pith_short_8","alias_value":"JZ7Y2DXF","created_at":"2026-07-17T00:20:57.977932+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/JZ7Y2DXFLTJZOYEAYCTTQVOUGH","json":"https://pith.science/pith/JZ7Y2DXFLTJZOYEAYCTTQVOUGH.json","graph_json":"https://pith.science/api/pith-number/JZ7Y2DXFLTJZOYEAYCTTQVOUGH/graph.json","events_json":"https://pith.science/api/pith-number/JZ7Y2DXFLTJZOYEAYCTTQVOUGH/events.json","paper":"https://pith.science/paper/JZ7Y2DXF"},"agent_actions":{"view_html":"https://pith.science/pith/JZ7Y2DXFLTJZOYEAYCTTQVOUGH","download_json":"https://pith.science/pith/JZ7Y2DXFLTJZOYEAYCTTQVOUGH.json","view_paper":"https://pith.science/paper/JZ7Y2DXF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.14205&json=true","fetch_graph":"https://pith.science/api/pith-number/JZ7Y2DXFLTJZOYEAYCTTQVOUGH/graph.json","fetch_events":"https://pith.science/api/pith-number/JZ7Y2DXFLTJZOYEAYCTTQVOUGH/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JZ7Y2DXFLTJZOYEAYCTTQVOUGH/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JZ7Y2DXFLTJZOYEAYCTTQVOUGH/action/storage_attestation","attest_author":"https://pith.science/pith/JZ7Y2DXFLTJZOYEAYCTTQVOUGH/action/author_attestation","sign_citation":"https://pith.science/pith/JZ7Y2DXFLTJZOYEAYCTTQVOUGH/action/citation_signature","submit_replication":"https://pith.science/pith/JZ7Y2DXFLTJZOYEAYCTTQVOUGH/action/replication_record"}},"created_at":"2026-07-17T00:20:57.977932+00:00","updated_at":"2026-07-17T00:20:57.977932+00:00"}