{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:S3FGFOCAUVS7336JBR3CDZLXSL","short_pith_number":"pith:S3FGFOCA","schema_version":"1.0","canonical_sha256":"96ca62b840a565fdefc90c7621e57792f0d5082f245145bf61b91a27065895b2","source":{"kind":"arxiv","id":"2103.06922","version":3},"attestation_state":"computed","paper":{"title":"Towards Interpreting and Mitigating Shortcut Learning Behavior of NLU Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Franck Dernoncourt, Jiuxiang Gu, Mengnan Du, Rajiv Jain, Ruchi Deshpande, Tong Sun, Varun Manjunatha, Xia Hu","submitted_at":"2021-03-11T19:39:56Z","abstract_excerpt":"Recent studies indicate that NLU models are prone to rely on shortcut features for prediction, without achieving true language understanding. As a result, these models fail to generalize to real-world out-of-distribution data. In this work, we show that the words in the NLU training set can be modeled as a long-tailed distribution. There are two findings: 1) NLU models have strong preference for features located at the head of the long-tailed distribution, and 2) Shortcut features are picked up during very early few iterations of the model training. These two observations are further employed "},"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":"2103.06922","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-03-11T19:39:56Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"2bace4d1414c41d1e8bb8da318e90791986a22e7c234211aec12c74d9eb3fc8f","abstract_canon_sha256":"277ef4fd57b5ea5791951322f18ff0f18a460e1da91966137c8e10c821ce82cf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:32:04.560051Z","signature_b64":"kB6iAyeINRtkBG0fyZ04JT1jgrW0KM1R4uGtGD1eaz/s83S072X8A+bMS2oF8uiXJME7AWfgF+Qcn+o0iLG+BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"96ca62b840a565fdefc90c7621e57792f0d5082f245145bf61b91a27065895b2","last_reissued_at":"2026-07-05T02:32:04.559556Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:32:04.559556Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Towards Interpreting and Mitigating Shortcut Learning Behavior of NLU Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Franck Dernoncourt, Jiuxiang Gu, Mengnan Du, Rajiv Jain, Ruchi Deshpande, Tong Sun, Varun Manjunatha, Xia Hu","submitted_at":"2021-03-11T19:39:56Z","abstract_excerpt":"Recent studies indicate that NLU models are prone to rely on shortcut features for prediction, without achieving true language understanding. As a result, these models fail to generalize to real-world out-of-distribution data. In this work, we show that the words in the NLU training set can be modeled as a long-tailed distribution. There are two findings: 1) NLU models have strong preference for features located at the head of the long-tailed distribution, and 2) Shortcut features are picked up during very early few iterations of the model training. These two observations are further employed "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2103.06922","kind":"arxiv","version":3},"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/2103.06922/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":"2103.06922","created_at":"2026-07-05T02:32:04.559637+00:00"},{"alias_kind":"arxiv_version","alias_value":"2103.06922v3","created_at":"2026-07-05T02:32:04.559637+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2103.06922","created_at":"2026-07-05T02:32:04.559637+00:00"},{"alias_kind":"pith_short_12","alias_value":"S3FGFOCAUVS7","created_at":"2026-07-05T02:32:04.559637+00:00"},{"alias_kind":"pith_short_16","alias_value":"S3FGFOCAUVS7336J","created_at":"2026-07-05T02:32:04.559637+00:00"},{"alias_kind":"pith_short_8","alias_value":"S3FGFOCA","created_at":"2026-07-05T02:32:04.559637+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.10930","citing_title":"Physics-informed Temporal Alignment for Auto-regressive PDE Foundation Models","ref_index":2006,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/S3FGFOCAUVS7336JBR3CDZLXSL","json":"https://pith.science/pith/S3FGFOCAUVS7336JBR3CDZLXSL.json","graph_json":"https://pith.science/api/pith-number/S3FGFOCAUVS7336JBR3CDZLXSL/graph.json","events_json":"https://pith.science/api/pith-number/S3FGFOCAUVS7336JBR3CDZLXSL/events.json","paper":"https://pith.science/paper/S3FGFOCA"},"agent_actions":{"view_html":"https://pith.science/pith/S3FGFOCAUVS7336JBR3CDZLXSL","download_json":"https://pith.science/pith/S3FGFOCAUVS7336JBR3CDZLXSL.json","view_paper":"https://pith.science/paper/S3FGFOCA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2103.06922&json=true","fetch_graph":"https://pith.science/api/pith-number/S3FGFOCAUVS7336JBR3CDZLXSL/graph.json","fetch_events":"https://pith.science/api/pith-number/S3FGFOCAUVS7336JBR3CDZLXSL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/S3FGFOCAUVS7336JBR3CDZLXSL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/S3FGFOCAUVS7336JBR3CDZLXSL/action/storage_attestation","attest_author":"https://pith.science/pith/S3FGFOCAUVS7336JBR3CDZLXSL/action/author_attestation","sign_citation":"https://pith.science/pith/S3FGFOCAUVS7336JBR3CDZLXSL/action/citation_signature","submit_replication":"https://pith.science/pith/S3FGFOCAUVS7336JBR3CDZLXSL/action/replication_record"}},"created_at":"2026-07-05T02:32:04.559637+00:00","updated_at":"2026-07-05T02:32:04.559637+00:00"}