{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:WWF7IPRFI3U5QRH3SLHHZ4EHDT","short_pith_number":"pith:WWF7IPRF","schema_version":"1.0","canonical_sha256":"b58bf43e2546e9d844fb92ce7cf0871cd7362efa87e286a7db7ee8be1ad7e4e8","source":{"kind":"arxiv","id":"2411.18674","version":2},"attestation_state":"computed","paper":{"title":"Active Data Curation Effectively Distills Large-Scale Multimodal Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Alessio Tonioni, Federico Tombari, Muhammad Ferjad Naeem, Nikhil Parthasarathy, Olivier J. H\\'enaff, Samuel Albanie, Talfan Evans, Vishaal Udandarao, Yongqin Xian","submitted_at":"2024-11-27T18:50:15Z","abstract_excerpt":"Knowledge distillation (KD) is the de facto standard for compressing large-scale models into smaller ones. Prior works have explored ever more complex KD strategies involving different objective functions, teacher-ensembles, and weight inheritance. In this work we explore an alternative, yet simple approach -- active data curation as effective distillation for contrastive multimodal pretraining. Our simple online batch selection method, ACID, outperforms strong KD baselines across various model-, data- and compute-configurations. Further, we find such an active data curation strategy to in fac"},"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":"2411.18674","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2024-11-27T18:50:15Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"bc1d6b0e6b3c29f92258057db88f456da9aded19227114f7546cd702308588d5","abstract_canon_sha256":"c4a9de80c678310788a4bcd864f70e12f28c1187c35ccb5d2a1d26a8cff85b09"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:58:20.780884Z","signature_b64":"V0oOEYUbK4ezgTsOUxF2DGD7QbhYabV7Qfb/2/X2MhQf3dJlbIaomTrkJiS8Bvi+CeMZ3jf393Fq52AU38MaDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b58bf43e2546e9d844fb92ce7cf0871cd7362efa87e286a7db7ee8be1ad7e4e8","last_reissued_at":"2026-07-05T10:58:20.780365Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:58:20.780365Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Active Data Curation Effectively Distills Large-Scale Multimodal Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CV","authors_text":"Alessio Tonioni, Federico Tombari, Muhammad Ferjad Naeem, Nikhil Parthasarathy, Olivier J. H\\'enaff, Samuel Albanie, Talfan Evans, Vishaal Udandarao, Yongqin Xian","submitted_at":"2024-11-27T18:50:15Z","abstract_excerpt":"Knowledge distillation (KD) is the de facto standard for compressing large-scale models into smaller ones. Prior works have explored ever more complex KD strategies involving different objective functions, teacher-ensembles, and weight inheritance. In this work we explore an alternative, yet simple approach -- active data curation as effective distillation for contrastive multimodal pretraining. Our simple online batch selection method, ACID, outperforms strong KD baselines across various model-, data- and compute-configurations. Further, we find such an active data curation strategy to in fac"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2411.18674","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/2411.18674/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":"2411.18674","created_at":"2026-07-05T10:58:20.780432+00:00"},{"alias_kind":"arxiv_version","alias_value":"2411.18674v2","created_at":"2026-07-05T10:58:20.780432+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2411.18674","created_at":"2026-07-05T10:58:20.780432+00:00"},{"alias_kind":"pith_short_12","alias_value":"WWF7IPRFI3U5","created_at":"2026-07-05T10:58:20.780432+00:00"},{"alias_kind":"pith_short_16","alias_value":"WWF7IPRFI3U5QRH3","created_at":"2026-07-05T10:58:20.780432+00:00"},{"alias_kind":"pith_short_8","alias_value":"WWF7IPRF","created_at":"2026-07-05T10:58:20.780432+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2502.14786","citing_title":"SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features","ref_index":61,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WWF7IPRFI3U5QRH3SLHHZ4EHDT","json":"https://pith.science/pith/WWF7IPRFI3U5QRH3SLHHZ4EHDT.json","graph_json":"https://pith.science/api/pith-number/WWF7IPRFI3U5QRH3SLHHZ4EHDT/graph.json","events_json":"https://pith.science/api/pith-number/WWF7IPRFI3U5QRH3SLHHZ4EHDT/events.json","paper":"https://pith.science/paper/WWF7IPRF"},"agent_actions":{"view_html":"https://pith.science/pith/WWF7IPRFI3U5QRH3SLHHZ4EHDT","download_json":"https://pith.science/pith/WWF7IPRFI3U5QRH3SLHHZ4EHDT.json","view_paper":"https://pith.science/paper/WWF7IPRF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2411.18674&json=true","fetch_graph":"https://pith.science/api/pith-number/WWF7IPRFI3U5QRH3SLHHZ4EHDT/graph.json","fetch_events":"https://pith.science/api/pith-number/WWF7IPRFI3U5QRH3SLHHZ4EHDT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WWF7IPRFI3U5QRH3SLHHZ4EHDT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WWF7IPRFI3U5QRH3SLHHZ4EHDT/action/storage_attestation","attest_author":"https://pith.science/pith/WWF7IPRFI3U5QRH3SLHHZ4EHDT/action/author_attestation","sign_citation":"https://pith.science/pith/WWF7IPRFI3U5QRH3SLHHZ4EHDT/action/citation_signature","submit_replication":"https://pith.science/pith/WWF7IPRFI3U5QRH3SLHHZ4EHDT/action/replication_record"}},"created_at":"2026-07-05T10:58:20.780432+00:00","updated_at":"2026-07-05T10:58:20.780432+00:00"}