{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:NTDTDZKB32E7CKTMHVJOBSOO5H","short_pith_number":"pith:NTDTDZKB","schema_version":"1.0","canonical_sha256":"6cc731e541de89f12a6c3d52e0c9cee9fb68ae6e8106f4146c2f4b3d2d15bc51","source":{"kind":"arxiv","id":"2309.17002","version":2},"attestation_state":"computed","paper":{"title":"Understanding and Mitigating the Label Noise in Pre-training on Downstream Tasks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Ankit Shah, Bhiksha Raj, Hao Chen, Hongxin Wei, Jindong Wang, Masashi Sugiyama, Ran Tao, Xing Xie","submitted_at":"2023-09-29T06:18:15Z","abstract_excerpt":"Pre-training on large-scale datasets and then fine-tuning on downstream tasks have become a standard practice in deep learning. However, pre-training data often contain label noise that may adversely affect the generalization of the model. This paper aims to understand the nature of noise in pre-training datasets and to mitigate its impact on downstream tasks. More specifically, through extensive experiments of supervised pre-training models on synthetic noisy ImageNet-1K and YFCC15M datasets, we demonstrate that while slight noise in pre-training can benefit in-domain (ID) transfer performanc"},"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":"2309.17002","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-09-29T06:18:15Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"5188987404c1ded245af936a4239f4433f3fd7736dfb2e83f6eae422c18984fa","abstract_canon_sha256":"106398e5ca7b0b0616a6223b6256f748a1fdc32592babf76cab96f85dc7b85de"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:54:33.130113Z","signature_b64":"ODtfC2mBwOcQvyHphFDPLju9nLHnmF4jrhI1hgKl+3z2yi6ge54yTMfgCW5z0mxj9ippiSAopaQHDLvHD42ZAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6cc731e541de89f12a6c3d52e0c9cee9fb68ae6e8106f4146c2f4b3d2d15bc51","last_reissued_at":"2026-07-05T07:54:33.129621Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:54:33.129621Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding and Mitigating the Label Noise in Pre-training on Downstream Tasks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Ankit Shah, Bhiksha Raj, Hao Chen, Hongxin Wei, Jindong Wang, Masashi Sugiyama, Ran Tao, Xing Xie","submitted_at":"2023-09-29T06:18:15Z","abstract_excerpt":"Pre-training on large-scale datasets and then fine-tuning on downstream tasks have become a standard practice in deep learning. However, pre-training data often contain label noise that may adversely affect the generalization of the model. This paper aims to understand the nature of noise in pre-training datasets and to mitigate its impact on downstream tasks. More specifically, through extensive experiments of supervised pre-training models on synthetic noisy ImageNet-1K and YFCC15M datasets, we demonstrate that while slight noise in pre-training can benefit in-domain (ID) transfer performanc"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2309.17002","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/2309.17002/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":"2309.17002","created_at":"2026-07-05T07:54:33.129677+00:00"},{"alias_kind":"arxiv_version","alias_value":"2309.17002v2","created_at":"2026-07-05T07:54:33.129677+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2309.17002","created_at":"2026-07-05T07:54:33.129677+00:00"},{"alias_kind":"pith_short_12","alias_value":"NTDTDZKB32E7","created_at":"2026-07-05T07:54:33.129677+00:00"},{"alias_kind":"pith_short_16","alias_value":"NTDTDZKB32E7CKTM","created_at":"2026-07-05T07:54:33.129677+00:00"},{"alias_kind":"pith_short_8","alias_value":"NTDTDZKB","created_at":"2026-07-05T07:54:33.129677+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2405.12969","citing_title":"EchoAlign: Bridging Generative and Discriminative Learning under Noisy Labels","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10129","citing_title":"Synthetic Pre-Pre-Training Improves Language Model Robustness to Noisy Pre-Training Data","ref_index":57,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NTDTDZKB32E7CKTMHVJOBSOO5H","json":"https://pith.science/pith/NTDTDZKB32E7CKTMHVJOBSOO5H.json","graph_json":"https://pith.science/api/pith-number/NTDTDZKB32E7CKTMHVJOBSOO5H/graph.json","events_json":"https://pith.science/api/pith-number/NTDTDZKB32E7CKTMHVJOBSOO5H/events.json","paper":"https://pith.science/paper/NTDTDZKB"},"agent_actions":{"view_html":"https://pith.science/pith/NTDTDZKB32E7CKTMHVJOBSOO5H","download_json":"https://pith.science/pith/NTDTDZKB32E7CKTMHVJOBSOO5H.json","view_paper":"https://pith.science/paper/NTDTDZKB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2309.17002&json=true","fetch_graph":"https://pith.science/api/pith-number/NTDTDZKB32E7CKTMHVJOBSOO5H/graph.json","fetch_events":"https://pith.science/api/pith-number/NTDTDZKB32E7CKTMHVJOBSOO5H/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NTDTDZKB32E7CKTMHVJOBSOO5H/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NTDTDZKB32E7CKTMHVJOBSOO5H/action/storage_attestation","attest_author":"https://pith.science/pith/NTDTDZKB32E7CKTMHVJOBSOO5H/action/author_attestation","sign_citation":"https://pith.science/pith/NTDTDZKB32E7CKTMHVJOBSOO5H/action/citation_signature","submit_replication":"https://pith.science/pith/NTDTDZKB32E7CKTMHVJOBSOO5H/action/replication_record"}},"created_at":"2026-07-05T07:54:33.129677+00:00","updated_at":"2026-07-05T07:54:33.129677+00:00"}