{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:PVLHFZZ4CFSZNFKLPIFTUU3CUA","short_pith_number":"pith:PVLHFZZ4","schema_version":"1.0","canonical_sha256":"7d5672e73c116596954b7a0b3a5362a02645eb012514c95f54d8deede2f1a291","source":{"kind":"arxiv","id":"2205.01397","version":2},"attestation_state":"computed","paper":{"title":"Data Determines Distributional Robustness in Contrastive Language Image Pre-training (CLIP)","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Achal Dave, Alex Fang, Gabriel Ilharco, Ludwig Schmidt, Mitchell Wortsman, Vaishaal Shankar, Yuhao Wan","submitted_at":"2022-05-03T10:06:51Z","abstract_excerpt":"Contrastively trained language-image models such as CLIP, ALIGN, and BASIC have demonstrated unprecedented robustness to multiple challenging natural distribution shifts. Since these language-image models differ from previous training approaches in several ways, an important question is what causes the large robustness gains. We answer this question via a systematic experimental investigation. Concretely, we study five different possible causes for the robustness gains: (i) the training set size, (ii) the training distribution, (iii) language supervision at training time, (iv) language supervi"},"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":"2205.01397","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2022-05-03T10:06:51Z","cross_cats_sorted":["cs.CL","cs.LG"],"title_canon_sha256":"797f08907b3eef373cd2f58bd090bf623da060c68f5428536a31149e4f3c1cb9","abstract_canon_sha256":"63e251200e93bdefea332f31000dc714b93781c39c8b36dafbca6b888786a934"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:50:34.287467Z","signature_b64":"sSMyfXDvs759vwShLNJR2jiuwVilccO29F81z1SEsQXSRUDjE2dTUuloD6HMNiNSQWsS/74mxZdUAKm/zl1hDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7d5672e73c116596954b7a0b3a5362a02645eb012514c95f54d8deede2f1a291","last_reissued_at":"2026-07-05T04:50:34.286996Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:50:34.286996Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Data Determines Distributional Robustness in Contrastive Language Image Pre-training (CLIP)","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","cs.LG"],"primary_cat":"cs.CV","authors_text":"Achal Dave, Alex Fang, Gabriel Ilharco, Ludwig Schmidt, Mitchell Wortsman, Vaishaal Shankar, Yuhao Wan","submitted_at":"2022-05-03T10:06:51Z","abstract_excerpt":"Contrastively trained language-image models such as CLIP, ALIGN, and BASIC have demonstrated unprecedented robustness to multiple challenging natural distribution shifts. Since these language-image models differ from previous training approaches in several ways, an important question is what causes the large robustness gains. We answer this question via a systematic experimental investigation. Concretely, we study five different possible causes for the robustness gains: (i) the training set size, (ii) the training distribution, (iii) language supervision at training time, (iv) language supervi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.01397","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/2205.01397/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":"2205.01397","created_at":"2026-07-05T04:50:34.287051+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.01397v2","created_at":"2026-07-05T04:50:34.287051+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.01397","created_at":"2026-07-05T04:50:34.287051+00:00"},{"alias_kind":"pith_short_12","alias_value":"PVLHFZZ4CFSZ","created_at":"2026-07-05T04:50:34.287051+00:00"},{"alias_kind":"pith_short_16","alias_value":"PVLHFZZ4CFSZNFKL","created_at":"2026-07-05T04:50:34.287051+00:00"},{"alias_kind":"pith_short_8","alias_value":"PVLHFZZ4","created_at":"2026-07-05T04:50:34.287051+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2502.12120","citing_title":"LLMs on the Line: Data Determines Loss-to-Loss Scaling Laws","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PVLHFZZ4CFSZNFKLPIFTUU3CUA","json":"https://pith.science/pith/PVLHFZZ4CFSZNFKLPIFTUU3CUA.json","graph_json":"https://pith.science/api/pith-number/PVLHFZZ4CFSZNFKLPIFTUU3CUA/graph.json","events_json":"https://pith.science/api/pith-number/PVLHFZZ4CFSZNFKLPIFTUU3CUA/events.json","paper":"https://pith.science/paper/PVLHFZZ4"},"agent_actions":{"view_html":"https://pith.science/pith/PVLHFZZ4CFSZNFKLPIFTUU3CUA","download_json":"https://pith.science/pith/PVLHFZZ4CFSZNFKLPIFTUU3CUA.json","view_paper":"https://pith.science/paper/PVLHFZZ4","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.01397&json=true","fetch_graph":"https://pith.science/api/pith-number/PVLHFZZ4CFSZNFKLPIFTUU3CUA/graph.json","fetch_events":"https://pith.science/api/pith-number/PVLHFZZ4CFSZNFKLPIFTUU3CUA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PVLHFZZ4CFSZNFKLPIFTUU3CUA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PVLHFZZ4CFSZNFKLPIFTUU3CUA/action/storage_attestation","attest_author":"https://pith.science/pith/PVLHFZZ4CFSZNFKLPIFTUU3CUA/action/author_attestation","sign_citation":"https://pith.science/pith/PVLHFZZ4CFSZNFKLPIFTUU3CUA/action/citation_signature","submit_replication":"https://pith.science/pith/PVLHFZZ4CFSZNFKLPIFTUU3CUA/action/replication_record"}},"created_at":"2026-07-05T04:50:34.287051+00:00","updated_at":"2026-07-05T04:50:34.287051+00:00"}