{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:3G4PXEQF54H54ICKVH25JAYLGC","short_pith_number":"pith:3G4PXEQF","schema_version":"1.0","canonical_sha256":"d9b8fb9205ef0fde204aa9f5d4830b308ed40621d25cd8b175bcba5e963719b8","source":{"kind":"arxiv","id":"2204.02937","version":2},"attestation_state":"computed","paper":{"title":"Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Andrew Gordon Wilson, Pavel Izmailov, Polina Kirichenko","submitted_at":"2022-04-06T16:55:41Z","abstract_excerpt":"Neural network classifiers can largely rely on simple spurious features, such as backgrounds, to make predictions. However, even in these cases, we show that they still often learn core features associated with the desired attributes of the data, contrary to recent findings. Inspired by this insight, we demonstrate that simple last layer retraining can match or outperform state-of-the-art approaches on spurious correlation benchmarks, but with profoundly lower complexity and computational expenses. Moreover, we show that last layer retraining on large ImageNet-trained models can also significa"},"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":"2204.02937","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2022-04-06T16:55:41Z","cross_cats_sorted":["cs.CV","stat.ML"],"title_canon_sha256":"ead46349c8ed5783299c86d9fddd1a1272f6192bd50f4538a1ebccf3deabbaa3","abstract_canon_sha256":"7e6d81acf03d2ca87e81dbb7d8f870c391299ecf158f9a04dc7e1fa4ca2ebdda"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:26:40.017494Z","signature_b64":"DuynU6Usd5jDUsNrFfTskoi2mItNLO30yxmcCz8s0ARcUsKL1FkaUhQSxPntS6jemwxDSnnHa7MuMd3gJC64DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d9b8fb9205ef0fde204aa9f5d4830b308ed40621d25cd8b175bcba5e963719b8","last_reissued_at":"2026-07-05T06:26:40.017030Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:26:40.017030Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Last Layer Re-Training is Sufficient for Robustness to Spurious Correlations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CV","stat.ML"],"primary_cat":"cs.LG","authors_text":"Andrew Gordon Wilson, Pavel Izmailov, Polina Kirichenko","submitted_at":"2022-04-06T16:55:41Z","abstract_excerpt":"Neural network classifiers can largely rely on simple spurious features, such as backgrounds, to make predictions. However, even in these cases, we show that they still often learn core features associated with the desired attributes of the data, contrary to recent findings. Inspired by this insight, we demonstrate that simple last layer retraining can match or outperform state-of-the-art approaches on spurious correlation benchmarks, but with profoundly lower complexity and computational expenses. Moreover, we show that last layer retraining on large ImageNet-trained models can also significa"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2204.02937","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/2204.02937/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":"2204.02937","created_at":"2026-07-05T06:26:40.017093+00:00"},{"alias_kind":"arxiv_version","alias_value":"2204.02937v2","created_at":"2026-07-05T06:26:40.017093+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2204.02937","created_at":"2026-07-05T06:26:40.017093+00:00"},{"alias_kind":"pith_short_12","alias_value":"3G4PXEQF54H5","created_at":"2026-07-05T06:26:40.017093+00:00"},{"alias_kind":"pith_short_16","alias_value":"3G4PXEQF54H54ICK","created_at":"2026-07-05T06:26:40.017093+00:00"},{"alias_kind":"pith_short_8","alias_value":"3G4PXEQF","created_at":"2026-07-05T06:26:40.017093+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":10,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.24161","citing_title":"Dual-Branch Cross-Projection Debiasing through Diffusion-based Disentanglement","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2606.18571","citing_title":"Fair Cognitive Impairment Detection Through Unlearning","ref_index":54,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03148","citing_title":"$A^2$: Smaller Self-Supervised ViTs Localize Better than Larger Ones","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2606.02830","citing_title":"Mitigating Spurious Correlations with Memorization-Guided Dataset De-Biasing","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.21602","citing_title":"Benchmarking and Improving Monitors for Out-Of-Distribution Alignment Failure in LLMs","ref_index":49,"is_internal_anchor":false},{"citing_arxiv_id":"2602.18502","citing_title":"Mitigating Shortcut Learning via Feature Disentanglement in Medical Imaging: A Benchmark Study","ref_index":33,"is_internal_anchor":false},{"citing_arxiv_id":"2403.19647","citing_title":"Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11107","citing_title":"Birds of a Feather Flock Together: Background-Invariant Representations via Linear Structure in VLMs","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11134","citing_title":"Spurious Correlation Learning in Preference Optimization: Mechanisms, Consequences, and Mitigation via Tie Training","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11237","citing_title":"DeconDTN-Toolkit: A Library for Evaluation and Enhancement of Robustness to Provenance Shift","ref_index":46,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3G4PXEQF54H54ICKVH25JAYLGC","json":"https://pith.science/pith/3G4PXEQF54H54ICKVH25JAYLGC.json","graph_json":"https://pith.science/api/pith-number/3G4PXEQF54H54ICKVH25JAYLGC/graph.json","events_json":"https://pith.science/api/pith-number/3G4PXEQF54H54ICKVH25JAYLGC/events.json","paper":"https://pith.science/paper/3G4PXEQF"},"agent_actions":{"view_html":"https://pith.science/pith/3G4PXEQF54H54ICKVH25JAYLGC","download_json":"https://pith.science/pith/3G4PXEQF54H54ICKVH25JAYLGC.json","view_paper":"https://pith.science/paper/3G4PXEQF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2204.02937&json=true","fetch_graph":"https://pith.science/api/pith-number/3G4PXEQF54H54ICKVH25JAYLGC/graph.json","fetch_events":"https://pith.science/api/pith-number/3G4PXEQF54H54ICKVH25JAYLGC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3G4PXEQF54H54ICKVH25JAYLGC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3G4PXEQF54H54ICKVH25JAYLGC/action/storage_attestation","attest_author":"https://pith.science/pith/3G4PXEQF54H54ICKVH25JAYLGC/action/author_attestation","sign_citation":"https://pith.science/pith/3G4PXEQF54H54ICKVH25JAYLGC/action/citation_signature","submit_replication":"https://pith.science/pith/3G4PXEQF54H54ICKVH25JAYLGC/action/replication_record"}},"created_at":"2026-07-05T06:26:40.017093+00:00","updated_at":"2026-07-05T06:26:40.017093+00:00"}