{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:I2432NZGDQ64IUVEVBDEKIFJKG","short_pith_number":"pith:I2432NZG","schema_version":"1.0","canonical_sha256":"46b9bd37261c3dc452a4a8464520a951a2c38869bc496956f3981f672e562e95","source":{"kind":"arxiv","id":"2105.11447","version":1},"attestation_state":"computed","paper":{"title":"True Few-Shot Learning with Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.CL","authors_text":"Douwe Kiela, Ethan Perez, Kyunghyun Cho","submitted_at":"2021-05-24T17:55:51Z","abstract_excerpt":"Pretrained language models (LMs) perform well on many tasks even when learning from a few examples, but prior work uses many held-out examples to tune various aspects of learning, such as hyperparameters, training objectives, and natural language templates (\"prompts\"). Here, we evaluate the few-shot ability of LMs when such held-out examples are unavailable, a setting we call true few-shot learning. We test two model selection criteria, cross-validation and minimum description length, for choosing LM prompts and hyperparameters in the true few-shot setting. On average, both marginally outperfo"},"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":"2105.11447","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-05-24T17:55:51Z","cross_cats_sorted":["cs.LG","stat.ML"],"title_canon_sha256":"2a20383752954009624d58367352df331e62f890827c466bb908849651ebaede","abstract_canon_sha256":"b69039803ced6479eab9427ad2f4d357e5c00fc4f6bf8172b84d7f7656daac9a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:42:38.504064Z","signature_b64":"U9VyVb3TAQULojLkwL3Ct24Z+JLnhaDlvuqJqA+gAZGuVwietR9Mp9TKUq9kx0nNW+AH0+9MsEm1PF5knQskBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"46b9bd37261c3dc452a4a8464520a951a2c38869bc496956f3981f672e562e95","last_reissued_at":"2026-07-05T02:42:38.503393Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:42:38.503393Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"True Few-Shot Learning with Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG","stat.ML"],"primary_cat":"cs.CL","authors_text":"Douwe Kiela, Ethan Perez, Kyunghyun Cho","submitted_at":"2021-05-24T17:55:51Z","abstract_excerpt":"Pretrained language models (LMs) perform well on many tasks even when learning from a few examples, but prior work uses many held-out examples to tune various aspects of learning, such as hyperparameters, training objectives, and natural language templates (\"prompts\"). Here, we evaluate the few-shot ability of LMs when such held-out examples are unavailable, a setting we call true few-shot learning. We test two model selection criteria, cross-validation and minimum description length, for choosing LM prompts and hyperparameters in the true few-shot setting. On average, both marginally outperfo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2105.11447","kind":"arxiv","version":1},"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/2105.11447/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":"2105.11447","created_at":"2026-07-05T02:42:38.503646+00:00"},{"alias_kind":"arxiv_version","alias_value":"2105.11447v1","created_at":"2026-07-05T02:42:38.503646+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2105.11447","created_at":"2026-07-05T02:42:38.503646+00:00"},{"alias_kind":"pith_short_12","alias_value":"I2432NZGDQ64","created_at":"2026-07-05T02:42:38.503646+00:00"},{"alias_kind":"pith_short_16","alias_value":"I2432NZGDQ64IUVE","created_at":"2026-07-05T02:42:38.503646+00:00"},{"alias_kind":"pith_short_8","alias_value":"I2432NZG","created_at":"2026-07-05T02:42:38.503646+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2202.12837","citing_title":"Rethinking the Role of Demonstrations: What Makes In-Context Learning Work?","ref_index":116,"is_internal_anchor":false},{"citing_arxiv_id":"2110.08207","citing_title":"Multitask Prompted Training Enables Zero-Shot Task Generalization","ref_index":42,"is_internal_anchor":false},{"citing_arxiv_id":"2406.06592","citing_title":"Improve Mathematical Reasoning in Language Models by Automated Process Supervision","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2109.07958","citing_title":"TruthfulQA: Measuring How Models Mimic Human Falsehoods","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2205.01068","citing_title":"OPT: Open Pre-trained Transformer Language Models","ref_index":207,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/I2432NZGDQ64IUVEVBDEKIFJKG","json":"https://pith.science/pith/I2432NZGDQ64IUVEVBDEKIFJKG.json","graph_json":"https://pith.science/api/pith-number/I2432NZGDQ64IUVEVBDEKIFJKG/graph.json","events_json":"https://pith.science/api/pith-number/I2432NZGDQ64IUVEVBDEKIFJKG/events.json","paper":"https://pith.science/paper/I2432NZG"},"agent_actions":{"view_html":"https://pith.science/pith/I2432NZGDQ64IUVEVBDEKIFJKG","download_json":"https://pith.science/pith/I2432NZGDQ64IUVEVBDEKIFJKG.json","view_paper":"https://pith.science/paper/I2432NZG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2105.11447&json=true","fetch_graph":"https://pith.science/api/pith-number/I2432NZGDQ64IUVEVBDEKIFJKG/graph.json","fetch_events":"https://pith.science/api/pith-number/I2432NZGDQ64IUVEVBDEKIFJKG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/I2432NZGDQ64IUVEVBDEKIFJKG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/I2432NZGDQ64IUVEVBDEKIFJKG/action/storage_attestation","attest_author":"https://pith.science/pith/I2432NZGDQ64IUVEVBDEKIFJKG/action/author_attestation","sign_citation":"https://pith.science/pith/I2432NZGDQ64IUVEVBDEKIFJKG/action/citation_signature","submit_replication":"https://pith.science/pith/I2432NZGDQ64IUVEVBDEKIFJKG/action/replication_record"}},"created_at":"2026-07-05T02:42:38.503646+00:00","updated_at":"2026-07-05T02:42:38.503646+00:00"}