{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:Y57DAC2RCP2JUGP75YMG4FIWIO","short_pith_number":"pith:Y57DAC2R","schema_version":"1.0","canonical_sha256":"c77e300b5113f49a19ffee186e151643b632aea22081a988c6dcd224c6f01bbe","source":{"kind":"arxiv","id":"2503.01208","version":2},"attestation_state":"computed","paper":{"title":"Watch Out Your Album! On the Inadvertent Privacy Memorization in Multi-Modal Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Gongshen Liu, Haodong Zhao, Hao Fei, Mong-Li Lee, Tianjie Ju, Wynne Hsu, Yi Hua, Yubin Zheng, Zhenyu Shao, Zhuosheng Zhang","submitted_at":"2025-03-03T06:10:27Z","abstract_excerpt":"Multi-Modal Large Language Models (MLLMs) have exhibited remarkable performance on various vision-language tasks such as Visual Question Answering (VQA). Despite accumulating evidence of privacy concerns associated with task-relevant content, it remains unclear whether MLLMs inadvertently memorize private content that is entirely irrelevant to the training tasks. In this paper, we investigate how randomly generated task-irrelevant private content can become spuriously correlated with downstream objectives due to partial mini-batch training dynamics, thus causing inadvertent memorization. Concr"},"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":"2503.01208","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-03-03T06:10:27Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"eeeaefb23d2d7cdf014020b1713b21befa004bb29e8c824bae35eb7a71167c39","abstract_canon_sha256":"fe930b0213b112784f9778336a158c4ddb912a34d3e1797a347cc75b1f043f68"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:21:43.556486Z","signature_b64":"RruvgFk9sQXkcSkYADbno366N4ub10E3koIofM3ySfY+4xFOXfOCNAiSsIAURceW/NHURddIGRwxHGmQ65QuAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c77e300b5113f49a19ffee186e151643b632aea22081a988c6dcd224c6f01bbe","last_reissued_at":"2026-07-05T11:21:43.555944Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:21:43.555944Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Watch Out Your Album! On the Inadvertent Privacy Memorization in Multi-Modal Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.CV","authors_text":"Gongshen Liu, Haodong Zhao, Hao Fei, Mong-Li Lee, Tianjie Ju, Wynne Hsu, Yi Hua, Yubin Zheng, Zhenyu Shao, Zhuosheng Zhang","submitted_at":"2025-03-03T06:10:27Z","abstract_excerpt":"Multi-Modal Large Language Models (MLLMs) have exhibited remarkable performance on various vision-language tasks such as Visual Question Answering (VQA). Despite accumulating evidence of privacy concerns associated with task-relevant content, it remains unclear whether MLLMs inadvertently memorize private content that is entirely irrelevant to the training tasks. In this paper, we investigate how randomly generated task-irrelevant private content can become spuriously correlated with downstream objectives due to partial mini-batch training dynamics, thus causing inadvertent memorization. Concr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.01208","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/2503.01208/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":"2503.01208","created_at":"2026-07-05T11:21:43.556003+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.01208v2","created_at":"2026-07-05T11:21:43.556003+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.01208","created_at":"2026-07-05T11:21:43.556003+00:00"},{"alias_kind":"pith_short_12","alias_value":"Y57DAC2RCP2J","created_at":"2026-07-05T11:21:43.556003+00:00"},{"alias_kind":"pith_short_16","alias_value":"Y57DAC2RCP2JUGP7","created_at":"2026-07-05T11:21:43.556003+00:00"},{"alias_kind":"pith_short_8","alias_value":"Y57DAC2R","created_at":"2026-07-05T11:21:43.556003+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.18455","citing_title":"OrganicHAR: Towards Activity Discovery in Organic Settings for Privacy Preserving Sensors Using Efficient Video Analysis","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2507.10610","citing_title":"LaSM: Layer-wise Scaling Mechanism for Defending Pop-up Attack on GUI Agents","ref_index":27,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Y57DAC2RCP2JUGP75YMG4FIWIO","json":"https://pith.science/pith/Y57DAC2RCP2JUGP75YMG4FIWIO.json","graph_json":"https://pith.science/api/pith-number/Y57DAC2RCP2JUGP75YMG4FIWIO/graph.json","events_json":"https://pith.science/api/pith-number/Y57DAC2RCP2JUGP75YMG4FIWIO/events.json","paper":"https://pith.science/paper/Y57DAC2R"},"agent_actions":{"view_html":"https://pith.science/pith/Y57DAC2RCP2JUGP75YMG4FIWIO","download_json":"https://pith.science/pith/Y57DAC2RCP2JUGP75YMG4FIWIO.json","view_paper":"https://pith.science/paper/Y57DAC2R","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.01208&json=true","fetch_graph":"https://pith.science/api/pith-number/Y57DAC2RCP2JUGP75YMG4FIWIO/graph.json","fetch_events":"https://pith.science/api/pith-number/Y57DAC2RCP2JUGP75YMG4FIWIO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Y57DAC2RCP2JUGP75YMG4FIWIO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Y57DAC2RCP2JUGP75YMG4FIWIO/action/storage_attestation","attest_author":"https://pith.science/pith/Y57DAC2RCP2JUGP75YMG4FIWIO/action/author_attestation","sign_citation":"https://pith.science/pith/Y57DAC2RCP2JUGP75YMG4FIWIO/action/citation_signature","submit_replication":"https://pith.science/pith/Y57DAC2RCP2JUGP75YMG4FIWIO/action/replication_record"}},"created_at":"2026-07-05T11:21:43.556003+00:00","updated_at":"2026-07-05T11:21:43.556003+00:00"}