{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:YER45XPBBQFOA2GSISPGBYIF45","short_pith_number":"pith:YER45XPB","schema_version":"1.0","canonical_sha256":"c123cedde10c0ae068d2449e60e105e7558572c75a73cc6201f3b070a37e64eb","source":{"kind":"arxiv","id":"2605.15460","version":1},"attestation_state":"computed","paper":{"title":"Differentially Private Motif-Preserving Multi-modal Hashing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Degree clipping before noisy synthesis lets differentially private multi-modal hashing retain up to 92.5 percent of non-private retrieval accuracy.","cross_cats":["cs.AI"],"primary_cat":"cs.IR","authors_text":"Jiahao Sun, Wei Dai, Zehua Cheng","submitted_at":"2026-05-14T22:43:30Z","abstract_excerpt":"Cross-modal hashing enables efficient retrieval by encoding images and text into compact binary codes. State-of-the-art methods rely on semantic similarity graphs derived from user interactions for supervision, yet these graphs encode sensitive behavioral patterns vulnerable to link reconstruction attacks. Existing privacy-preserving approaches fail on graph-structured data: Differentially Private SGD destroys relational motifs by treating samples independently, while graph synthesis methods suffer from unbounded local sensitivity in scale-free networks, hub nodes cause single-edge modificatio"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.15460","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.IR","submitted_at":"2026-05-14T22:43:30Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"4a18356291197261c8689a35ef5c251e3318b10835fcc221d96182ce65e48f61","abstract_canon_sha256":"62263ae63dafad58291d982776f6e1a16bea20e2c13717e7f870bce567e892be"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:00:59.721475Z","signature_b64":"yhYfNKem1pHc7EZ2pcQFHMsfhcFuaHgCnYOmpUjRxWIpyO3L80jv97ww3pkE6aosqszxHGCjB3xbtI6AwpsiDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c123cedde10c0ae068d2449e60e105e7558572c75a73cc6201f3b070a37e64eb","last_reissued_at":"2026-05-20T00:00:59.720643Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:00:59.720643Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Differentially Private Motif-Preserving Multi-modal Hashing","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Degree clipping before noisy synthesis lets differentially private multi-modal hashing retain up to 92.5 percent of non-private retrieval accuracy.","cross_cats":["cs.AI"],"primary_cat":"cs.IR","authors_text":"Jiahao Sun, Wei Dai, Zehua Cheng","submitted_at":"2026-05-14T22:43:30Z","abstract_excerpt":"Cross-modal hashing enables efficient retrieval by encoding images and text into compact binary codes. State-of-the-art methods rely on semantic similarity graphs derived from user interactions for supervision, yet these graphs encode sensitive behavioral patterns vulnerable to link reconstruction attacks. Existing privacy-preserving approaches fail on graph-structured data: Differentially Private SGD destroys relational motifs by treating samples independently, while graph synthesis methods suffer from unbounded local sensitivity in scale-free networks, hub nodes cause single-edge modificatio"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"DMP-MH outperforms private baselines by up to 11.4 mAP points while retaining up to 92.5% of non-private performance on MIRFlickr-25K and NUS-WIDE under a strict inductive protocol.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That deterministically clipping node degrees to a fixed threshold caps the L2-sensitivity of triangle motifs independently of dataset size and still leaves enough structural signal for the downstream hashing loss to recover useful cross-modal alignments.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DMP-MH clips degrees to control triangle sensitivity, synthesizes an edge-DP graph with Noisy Mirror Descent, and distills it into dual-stream hash networks, beating private baselines by up to 11.4 mAP on MIRFlickr-25K and NUS-WIDE while keeping 92.5% of non-private performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Degree clipping before noisy synthesis lets differentially private multi-modal hashing retain up to 92.5 percent of non-private retrieval accuracy.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"ce1da16fdb504972485685a17f04fceee3da3b5d5de010b9948395866451d9c0"},"source":{"id":"2605.15460","kind":"arxiv","version":1},"verdict":{"id":"84827b2f-a438-493b-8e96-805084abf309","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T14:51:28.525043Z","strongest_claim":"DMP-MH outperforms private baselines by up to 11.4 mAP points while retaining up to 92.5% of non-private performance on MIRFlickr-25K and NUS-WIDE under a strict inductive protocol.","one_line_summary":"DMP-MH clips degrees to control triangle sensitivity, synthesizes an edge-DP graph with Noisy Mirror Descent, and distills it into dual-stream hash networks, beating private baselines by up to 11.4 mAP on MIRFlickr-25K and NUS-WIDE while keeping 92.5% of non-private performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That deterministically clipping node degrees to a fixed threshold caps the L2-sensitivity of triangle motifs independently of dataset size and still leaves enough structural signal for the downstream hashing loss to recover useful cross-modal alignments.","pith_extraction_headline":"Degree clipping before noisy synthesis lets differentially private multi-modal hashing retain up to 92.5 percent of non-private retrieval accuracy."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.15460/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"citation_quote_validity","ran_at":"2026-05-19T15:49:45.738120Z","status":"completed","version":"0.1.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T15:23:28.905635Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T15:01:17.604419Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T15:00:38.907785Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T14:21:54.098784Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.671046Z","status":"skipped","version":"1.0.0","findings_count":0}],"snapshot_sha256":"6a8459b9fcbd0cdd56d40a23a17bd7b1b2bf49e3466d9ea4e9a0c015b8fe53df"},"references":{"count":43,"sample":[{"doi":"","year":2016,"title":"Martin Abadi, Andy Chu, Ian Goodfellow, H Brendan McMahan, Ilya Mironov, Kunal Talwar, and Li Zhang. 2016. Deep learning with differential privacy. In Proceedings of the 2016 ACM SIGSAC conference on ","work_id":"c9059fb9-6d33-40b8-a53f-cdec0e3f03eb","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Tadas Baltrušaitis, Chaitanya Ahuja, and Louis-Philippe Morency. 2018. Multi- modal machine learning: A survey and taxonomy.IEEE transactions on pattern analysis and machine intelligence41, 2 (2018), ","work_id":"454c840f-6b47-4250-a841-76bb7db96606","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1999,"title":"Albert-László Barabási and Réka Albert. 1999. Emergence of scaling in random networks.science286, 5439 (1999), 509–512","work_id":"6bd5dcd3-9949-418d-8f56-ebb9b91ff1c2","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2016,"title":"Austin R Benson, David F Gleich, and Jure Leskovec. 2016. Higher-order organi- zation of complex networks.Science353, 6295 (2016), 163–166","work_id":"b790660d-e4b4-474c-b1f2-78dfa0943aba","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1999,"title":"Dimitri P. Bertsekas. 1999.Nonlinear Programming(2 ed.). Athena Scientific","work_id":"d7bd4363-7c65-40e8-bc19-3d169066edd6","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":43,"snapshot_sha256":"816ae73e29e2a2e4796f237e4892cb03cfaa62e789ceb3da8e225f4b24fd12b1","internal_anchors":5},"formal_canon":{"evidence_count":2,"snapshot_sha256":"aff0a6b49c4f9799e365224268e0e61d70dc22050fad5ec07625766f10eeb7d6"},"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":"2605.15460","created_at":"2026-05-20T00:00:59.720786+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.15460v1","created_at":"2026-05-20T00:00:59.720786+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15460","created_at":"2026-05-20T00:00:59.720786+00:00"},{"alias_kind":"pith_short_12","alias_value":"YER45XPBBQFO","created_at":"2026-05-20T00:00:59.720786+00:00"},{"alias_kind":"pith_short_16","alias_value":"YER45XPBBQFOA2GS","created_at":"2026-05-20T00:00:59.720786+00:00"},{"alias_kind":"pith_short_8","alias_value":"YER45XPB","created_at":"2026-05-20T00:00:59.720786+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YER45XPBBQFOA2GSISPGBYIF45","json":"https://pith.science/pith/YER45XPBBQFOA2GSISPGBYIF45.json","graph_json":"https://pith.science/api/pith-number/YER45XPBBQFOA2GSISPGBYIF45/graph.json","events_json":"https://pith.science/api/pith-number/YER45XPBBQFOA2GSISPGBYIF45/events.json","paper":"https://pith.science/paper/YER45XPB"},"agent_actions":{"view_html":"https://pith.science/pith/YER45XPBBQFOA2GSISPGBYIF45","download_json":"https://pith.science/pith/YER45XPBBQFOA2GSISPGBYIF45.json","view_paper":"https://pith.science/paper/YER45XPB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.15460&json=true","fetch_graph":"https://pith.science/api/pith-number/YER45XPBBQFOA2GSISPGBYIF45/graph.json","fetch_events":"https://pith.science/api/pith-number/YER45XPBBQFOA2GSISPGBYIF45/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YER45XPBBQFOA2GSISPGBYIF45/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YER45XPBBQFOA2GSISPGBYIF45/action/storage_attestation","attest_author":"https://pith.science/pith/YER45XPBBQFOA2GSISPGBYIF45/action/author_attestation","sign_citation":"https://pith.science/pith/YER45XPBBQFOA2GSISPGBYIF45/action/citation_signature","submit_replication":"https://pith.science/pith/YER45XPBBQFOA2GSISPGBYIF45/action/replication_record"}},"created_at":"2026-05-20T00:00:59.720786+00:00","updated_at":"2026-05-20T00:00:59.720786+00:00"}