{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:REZHTCQRCZV3HKO2Z3HS65PBM3","short_pith_number":"pith:REZHTCQR","schema_version":"1.0","canonical_sha256":"8932798a11166bb3a9dacecf2f75e166e006bffb064d2179d0331e3e1027c093","source":{"kind":"arxiv","id":"2103.14749","version":4},"attestation_state":"computed","paper":{"title":"Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Anish Athalye, Curtis G. Northcutt, Jonas Mueller","submitted_at":"2021-03-26T21:54:36Z","abstract_excerpt":"We identify label errors in the test sets of 10 of the most commonly-used computer vision, natural language, and audio datasets, and subsequently study the potential for these label errors to affect benchmark results. Errors in test sets are numerous and widespread: we estimate an average of at least 3.3% errors across the 10 datasets, where for example label errors comprise at least 6% of the ImageNet validation set. Putative label errors are identified using confident learning algorithms and then human-validated via crowdsourcing (51% of the algorithmically-flagged candidates are indeed erro"},"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":"2103.14749","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2021-03-26T21:54:36Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"a82fc7eaba50cfd80027c2ffd6b1791163bfa9ed5fc72d35062d926924beb8e1","abstract_canon_sha256":"1cf5eb8f9d4a07e1f5044d84d9a113ae71e374b45bc1bbd4e2c06095d46b9646"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:29:48.043663Z","signature_b64":"8nXyspMMxYHzU7gDKz+XaR2As7bWSQ0XjEUi1r51zJvQOXcrAHMD0qxN/rDLiYZH03+vqLGez1498wCDW7/gDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8932798a11166bb3a9dacecf2f75e166e006bffb064d2179d0331e3e1027c093","last_reissued_at":"2026-07-05T03:29:48.043233Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:29:48.043233Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"stat.ML","authors_text":"Anish Athalye, Curtis G. Northcutt, Jonas Mueller","submitted_at":"2021-03-26T21:54:36Z","abstract_excerpt":"We identify label errors in the test sets of 10 of the most commonly-used computer vision, natural language, and audio datasets, and subsequently study the potential for these label errors to affect benchmark results. Errors in test sets are numerous and widespread: we estimate an average of at least 3.3% errors across the 10 datasets, where for example label errors comprise at least 6% of the ImageNet validation set. Putative label errors are identified using confident learning algorithms and then human-validated via crowdsourcing (51% of the algorithmically-flagged candidates are indeed erro"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2103.14749","kind":"arxiv","version":4},"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/2103.14749/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":"2103.14749","created_at":"2026-07-05T03:29:48.043298+00:00"},{"alias_kind":"arxiv_version","alias_value":"2103.14749v4","created_at":"2026-07-05T03:29:48.043298+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2103.14749","created_at":"2026-07-05T03:29:48.043298+00:00"},{"alias_kind":"pith_short_12","alias_value":"REZHTCQRCZV3","created_at":"2026-07-05T03:29:48.043298+00:00"},{"alias_kind":"pith_short_16","alias_value":"REZHTCQRCZV3HKO2","created_at":"2026-07-05T03:29:48.043298+00:00"},{"alias_kind":"pith_short_8","alias_value":"REZHTCQR","created_at":"2026-07-05T03:29:48.043298+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22437","citing_title":"MMGist: A Comprehensive Multimodal Benchmark for 2027","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19186","citing_title":"Learning to Annotate Delayed and False AEB Events: A Practical System for Extreme Class Imbalance and Asymmetric Label Noise","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.26973","citing_title":"Signal-to-Noise Ratio and Sample Size Govern Representational Alignment in Neural Networks","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2605.29720","citing_title":"Efficient, Validation-Free Intrinsic Quality Estimation for Large-Scale Face Recognition Datasets","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2405.00892","citing_title":"Wake Vision: A Tailored Dataset and Benchmark Suite for TinyML Computer Vision Applications","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2511.02627","citing_title":"DecompSR: A dataset for decomposed analyses of compositional multihop spatial reasoning","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2411.04872","citing_title":"FrontierMath: A Benchmark for Evaluating Advanced Mathematical Reasoning in AI","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2604.25923","citing_title":"Evaluation Revisited: A Taxonomy of Evaluation Concerns in Natural Language Processing","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2604.05329","citing_title":"Semantic Trimming and Auxiliary Multi-step Prediction for Generative Recommendation","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2604.17022","citing_title":"Beyond Black-Box Labels: Interpretable Criteria for Diagnosing Subjective NLP Tasks","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/REZHTCQRCZV3HKO2Z3HS65PBM3","json":"https://pith.science/pith/REZHTCQRCZV3HKO2Z3HS65PBM3.json","graph_json":"https://pith.science/api/pith-number/REZHTCQRCZV3HKO2Z3HS65PBM3/graph.json","events_json":"https://pith.science/api/pith-number/REZHTCQRCZV3HKO2Z3HS65PBM3/events.json","paper":"https://pith.science/paper/REZHTCQR"},"agent_actions":{"view_html":"https://pith.science/pith/REZHTCQRCZV3HKO2Z3HS65PBM3","download_json":"https://pith.science/pith/REZHTCQRCZV3HKO2Z3HS65PBM3.json","view_paper":"https://pith.science/paper/REZHTCQR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2103.14749&json=true","fetch_graph":"https://pith.science/api/pith-number/REZHTCQRCZV3HKO2Z3HS65PBM3/graph.json","fetch_events":"https://pith.science/api/pith-number/REZHTCQRCZV3HKO2Z3HS65PBM3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/REZHTCQRCZV3HKO2Z3HS65PBM3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/REZHTCQRCZV3HKO2Z3HS65PBM3/action/storage_attestation","attest_author":"https://pith.science/pith/REZHTCQRCZV3HKO2Z3HS65PBM3/action/author_attestation","sign_citation":"https://pith.science/pith/REZHTCQRCZV3HKO2Z3HS65PBM3/action/citation_signature","submit_replication":"https://pith.science/pith/REZHTCQRCZV3HKO2Z3HS65PBM3/action/replication_record"}},"created_at":"2026-07-05T03:29:48.043298+00:00","updated_at":"2026-07-05T03:29:48.043298+00:00"}