{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:64XJLEMGD2GGGUQ6RHHZ5OUF4C","short_pith_number":"pith:64XJLEMG","schema_version":"1.0","canonical_sha256":"f72e9591861e8c63521e89cf9eba85e0be8b9f695fdfe4d82c6e92ad490905d7","source":{"kind":"arxiv","id":"2204.03632","version":2},"attestation_state":"computed","paper":{"title":"The Effects of Regularization and Data Augmentation are Class Dependent","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Leon Bottou, Randall Balestriero, Yann LeCun","submitted_at":"2022-04-07T17:57:29Z","abstract_excerpt":"Regularization is a fundamental technique to prevent over-fitting and to improve generalization performances by constraining a model's complexity. Current Deep Networks heavily rely on regularizers such as Data-Augmentation (DA) or weight-decay, and employ structural risk minimization, i.e. cross-validation, to select the optimal regularization hyper-parameters. In this study, we demonstrate that techniques such as DA or weight decay produce a model with a reduced complexity that is unfair across classes. The optimal amount of DA or weight decay found from cross-validation leads to disastrous "},"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":"2204.03632","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2022-04-07T17:57:29Z","cross_cats_sorted":["cs.CV","stat.ML"],"title_canon_sha256":"62d75e9f5bcf4d8ea0a51054b7f91334367dad5080398a9e2d02f795b210c145","abstract_canon_sha256":"17e24eeb6d89187375f643e1c7cfd6c251be605f95aefee2ca7147b8232b3aa1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:12:54.406485Z","signature_b64":"wVk5y4Pgd1KzQ45DsNT2FtbCIQNtGhx2MZlAGRJcdMc+eIxmT9P6V6EzZ8m1OPic5RD4jIYYZZOJKW5Z4/v0AQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f72e9591861e8c63521e89cf9eba85e0be8b9f695fdfe4d82c6e92ad490905d7","last_reissued_at":"2026-07-05T04:12:54.406047Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:12:54.406047Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Effects of Regularization and Data Augmentation are Class Dependent","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Leon Bottou, Randall Balestriero, Yann LeCun","submitted_at":"2022-04-07T17:57:29Z","abstract_excerpt":"Regularization is a fundamental technique to prevent over-fitting and to improve generalization performances by constraining a model's complexity. Current Deep Networks heavily rely on regularizers such as Data-Augmentation (DA) or weight-decay, and employ structural risk minimization, i.e. cross-validation, to select the optimal regularization hyper-parameters. In this study, we demonstrate that techniques such as DA or weight decay produce a model with a reduced complexity that is unfair across classes. The optimal amount of DA or weight decay found from cross-validation leads to disastrous "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2204.03632","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/2204.03632/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":"2204.03632","created_at":"2026-07-05T04:12:54.406104+00:00"},{"alias_kind":"arxiv_version","alias_value":"2204.03632v2","created_at":"2026-07-05T04:12:54.406104+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2204.03632","created_at":"2026-07-05T04:12:54.406104+00:00"},{"alias_kind":"pith_short_12","alias_value":"64XJLEMGD2GG","created_at":"2026-07-05T04:12:54.406104+00:00"},{"alias_kind":"pith_short_16","alias_value":"64XJLEMGD2GGGUQ6","created_at":"2026-07-05T04:12:54.406104+00:00"},{"alias_kind":"pith_short_8","alias_value":"64XJLEMG","created_at":"2026-07-05T04:12:54.406104+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2502.06831","citing_title":"No Location Left Behind: Measuring and Improving the Fairness of Implicit Representations for Earth Data","ref_index":4,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/64XJLEMGD2GGGUQ6RHHZ5OUF4C","json":"https://pith.science/pith/64XJLEMGD2GGGUQ6RHHZ5OUF4C.json","graph_json":"https://pith.science/api/pith-number/64XJLEMGD2GGGUQ6RHHZ5OUF4C/graph.json","events_json":"https://pith.science/api/pith-number/64XJLEMGD2GGGUQ6RHHZ5OUF4C/events.json","paper":"https://pith.science/paper/64XJLEMG"},"agent_actions":{"view_html":"https://pith.science/pith/64XJLEMGD2GGGUQ6RHHZ5OUF4C","download_json":"https://pith.science/pith/64XJLEMGD2GGGUQ6RHHZ5OUF4C.json","view_paper":"https://pith.science/paper/64XJLEMG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2204.03632&json=true","fetch_graph":"https://pith.science/api/pith-number/64XJLEMGD2GGGUQ6RHHZ5OUF4C/graph.json","fetch_events":"https://pith.science/api/pith-number/64XJLEMGD2GGGUQ6RHHZ5OUF4C/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/64XJLEMGD2GGGUQ6RHHZ5OUF4C/action/timestamp_anchor","attest_storage":"https://pith.science/pith/64XJLEMGD2GGGUQ6RHHZ5OUF4C/action/storage_attestation","attest_author":"https://pith.science/pith/64XJLEMGD2GGGUQ6RHHZ5OUF4C/action/author_attestation","sign_citation":"https://pith.science/pith/64XJLEMGD2GGGUQ6RHHZ5OUF4C/action/citation_signature","submit_replication":"https://pith.science/pith/64XJLEMGD2GGGUQ6RHHZ5OUF4C/action/replication_record"}},"created_at":"2026-07-05T04:12:54.406104+00:00","updated_at":"2026-07-05T04:12:54.406104+00:00"}