{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:ENPNPDUFLG3SVW2SPDNUVOI45Y","short_pith_number":"pith:ENPNPDUF","schema_version":"1.0","canonical_sha256":"235ed78e8559b72adb5278db4ab91cee060c48f122170203659932ad6699c3c6","source":{"kind":"arxiv","id":"2006.05987","version":3},"attestation_state":"computed","paper":{"title":"Revisiting Few-sample BERT Fine-tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Arzoo Katiyar, Felix Wu, Kilian Q. Weinberger, Tianyi Zhang, Yoav Artzi","submitted_at":"2020-06-10T17:57:03Z","abstract_excerpt":"This paper is a study of fine-tuning of BERT contextual representations, with focus on commonly observed instabilities in few-sample scenarios. We identify several factors that cause this instability: the common use of a non-standard optimization method with biased gradient estimation; the limited applicability of significant parts of the BERT network for down-stream tasks; and the prevalent practice of using a pre-determined, and small number of training iterations. We empirically test the impact of these factors, and identify alternative practices that resolve the commonly observed instabili"},"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":"2006.05987","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-06-10T17:57:03Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"6f84d9a748a391e5266bff9b91b8c524f34de7a8e67c6590b6b0e823ee4cf7b1","abstract_canon_sha256":"16db4500351188381b7761816997a8d5fe4ef1c0d0769b9916ca86197a3c3556"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:22:16.377809Z","signature_b64":"W+wPLxwiEOO8lp3Kp6Q0l4vj4fKDbuFrgp8LiGvE2iV2EdVsS8z30k8C66dGHO+0S/lHD+6GWZfwgJj0rYD9Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"235ed78e8559b72adb5278db4ab91cee060c48f122170203659932ad6699c3c6","last_reissued_at":"2026-07-05T02:22:16.377416Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:22:16.377416Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Revisiting Few-sample BERT Fine-tuning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Arzoo Katiyar, Felix Wu, Kilian Q. Weinberger, Tianyi Zhang, Yoav Artzi","submitted_at":"2020-06-10T17:57:03Z","abstract_excerpt":"This paper is a study of fine-tuning of BERT contextual representations, with focus on commonly observed instabilities in few-sample scenarios. We identify several factors that cause this instability: the common use of a non-standard optimization method with biased gradient estimation; the limited applicability of significant parts of the BERT network for down-stream tasks; and the prevalent practice of using a pre-determined, and small number of training iterations. We empirically test the impact of these factors, and identify alternative practices that resolve the commonly observed instabili"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2006.05987","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/2006.05987/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":"2006.05987","created_at":"2026-07-05T02:22:16.377474+00:00"},{"alias_kind":"arxiv_version","alias_value":"2006.05987v3","created_at":"2026-07-05T02:22:16.377474+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2006.05987","created_at":"2026-07-05T02:22:16.377474+00:00"},{"alias_kind":"pith_short_12","alias_value":"ENPNPDUFLG3S","created_at":"2026-07-05T02:22:16.377474+00:00"},{"alias_kind":"pith_short_16","alias_value":"ENPNPDUFLG3SVW2S","created_at":"2026-07-05T02:22:16.377474+00:00"},{"alias_kind":"pith_short_8","alias_value":"ENPNPDUF","created_at":"2026-07-05T02:22:16.377474+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22917","citing_title":"GRAIN: Group Aggregation via Min-Norm Objective","ref_index":80,"is_internal_anchor":false},{"citing_arxiv_id":"2507.00994","citing_title":"Should We Still Pretrain Encoders with Masked Language Modeling?","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2605.12516","citing_title":"Domain Adaptation of Large Language Models for Polymer-Composite Additive Manufacturing Using Retrieval-Augmented Generation and Fine-Tuning","ref_index":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ENPNPDUFLG3SVW2SPDNUVOI45Y","json":"https://pith.science/pith/ENPNPDUFLG3SVW2SPDNUVOI45Y.json","graph_json":"https://pith.science/api/pith-number/ENPNPDUFLG3SVW2SPDNUVOI45Y/graph.json","events_json":"https://pith.science/api/pith-number/ENPNPDUFLG3SVW2SPDNUVOI45Y/events.json","paper":"https://pith.science/paper/ENPNPDUF"},"agent_actions":{"view_html":"https://pith.science/pith/ENPNPDUFLG3SVW2SPDNUVOI45Y","download_json":"https://pith.science/pith/ENPNPDUFLG3SVW2SPDNUVOI45Y.json","view_paper":"https://pith.science/paper/ENPNPDUF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2006.05987&json=true","fetch_graph":"https://pith.science/api/pith-number/ENPNPDUFLG3SVW2SPDNUVOI45Y/graph.json","fetch_events":"https://pith.science/api/pith-number/ENPNPDUFLG3SVW2SPDNUVOI45Y/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ENPNPDUFLG3SVW2SPDNUVOI45Y/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ENPNPDUFLG3SVW2SPDNUVOI45Y/action/storage_attestation","attest_author":"https://pith.science/pith/ENPNPDUFLG3SVW2SPDNUVOI45Y/action/author_attestation","sign_citation":"https://pith.science/pith/ENPNPDUFLG3SVW2SPDNUVOI45Y/action/citation_signature","submit_replication":"https://pith.science/pith/ENPNPDUFLG3SVW2SPDNUVOI45Y/action/replication_record"}},"created_at":"2026-07-05T02:22:16.377474+00:00","updated_at":"2026-07-05T02:22:16.377474+00:00"}