{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2021:JDVYZVMEZKIKYMTR5Y7GAZ2ICM","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"0ba08acab8d5daef298eda40b969a5252fbad3b10cd5421f91a7fa217ca5a3b2","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-04-18T11:27:08Z","title_canon_sha256":"521eac38e6d56e5902f623741973ee63232a52df03edcbb497fbb9d6cc2c02e0"},"schema_version":"1.0","source":{"id":"2104.08821","kind":"arxiv","version":4}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2104.08821","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"arxiv_version","alias_value":"2104.08821v4","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2104.08821","created_at":"2026-05-17T23:38:53Z"},{"alias_kind":"pith_short_12","alias_value":"JDVYZVMEZKIK","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_16","alias_value":"JDVYZVMEZKIKYMTR","created_at":"2026-05-18T12:33:33Z"},{"alias_kind":"pith_short_8","alias_value":"JDVYZVME","created_at":"2026-05-18T12:33:33Z"}],"graph_snapshots":[{"event_id":"sha256:cad7e423ee5d9e2994556c2ebf6261ea77e785c591e4af547c7bb0a01c13a668","target":"graph","created_at":"2026-05-17T23:38:53Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"our unsupervised and supervised models using BERT base achieve an average of 76.3% and 81.6% Spearman's correlation respectively, a 4.2% and 2.2% improvement compared to the previous best results. We also show -- both theoretically and empirically -- that the contrastive learning objective regularizes pre-trained embeddings' anisotropic space to be more uniform"},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"that standard dropout is sufficient as data augmentation to prevent representation collapse in the unsupervised contrastive objective, and that NLI entailment/contradiction pairs form appropriate positive and hard-negative pairs for learning general sentence embeddings."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"SimCSE achieves 76.3% unsupervised and 81.6% supervised Spearman's correlation on STS tasks with BERT-base, improving prior best results by 4.2% and 2.2% via simple contrastive learning."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Contrastive learning with standard dropout as the only noise produces sentence embeddings that match or beat prior supervised results."}],"snapshot_sha256":"d23982583bab21067a7b361fe4fdb8ba245a7c8e32535fc7425b7316d1e1f02c"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"948452d7b3a30d0f4065d16892d7691db9b52fabea99a1d4f53ea8f85569c3da"},"paper":{"abstract_excerpt":"This paper presents SimCSE, a simple contrastive learning framework that greatly advances state-of-the-art sentence embeddings. We first describe an unsupervised approach, which takes an input sentence and predicts itself in a contrastive objective, with only standard dropout used as noise. This simple method works surprisingly well, performing on par with previous supervised counterparts. We find that dropout acts as minimal data augmentation, and removing it leads to a representation collapse. Then, we propose a supervised approach, which incorporates annotated pairs from natural language in","authors_text":"Danqi Chen, Tianyu Gao, Xingcheng Yao","cross_cats":["cs.LG"],"headline":"Contrastive learning with standard dropout as the only noise produces sentence embeddings that match or beat prior supervised results.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-04-18T11:27:08Z","title":"SimCSE: Simple Contrastive Learning of Sentence Embeddings"},"references":{"count":109,"internal_anchors":4,"resolved_work":109,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Eneko Agirre, Daniel Cer, Mona Diab, and Aitor Gonzalez-Agirre. 2012. https://www.aclweb.org/anthology/S12-1051 S em E val-2012 task 6: A pilot on semantic textual similarity . In * SEM 2012: The Firs","work_id":"5c5985e9-1814-48b7-bcb4-7729a1c061f6","year":2012},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Eneko Agirre, Daniel Cer, Mona Diab, Aitor Gonzalez-Agirre, and Weiwei Guo. 2013. https://www.aclweb.org/anthology/S13-1004 * SEM 2013 shared task: Semantic textual similarity . In Second Joint Confer","work_id":"bbbddcb5-789b-4d66-8a0e-ca0c3d63d8c3","year":2013},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":6,"title":"Sanjeev Arora, Yingyu Liang, and Tengyu Ma. 2017. https://openreview.net/forum?id=SyK00v5xx A simple but tough-to-beat baseline for sentence embeddings . In International Conference on Learning Repres","work_id":"134c4c07-53b7-4f57-a9fc-8cec09d1cfc2","year":2017},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":8,"title":"Fredrik Carlsson, Amaru Cuba Gyllensten, Evangelia Gogoulou, Erik Ylip \\\"a \\\"a Hellqvist, and Magnus Sahlgren. 2021. https://openreview.net/forum?id=Ov_sMNau-PF Semantic re-tuning with contrastive ten","work_id":"24bac75d-ff61-4dd9-9ec9-9902d815cc3f","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":11,"title":"Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. http://proceedings.mlr.press/v119/chen20j.html A simple framework for contrastive learning of visual representations . In Inter","work_id":"643fd409-2cee-4151-8cfb-7a1f881694a1","year":2020}],"snapshot_sha256":"da93f7ddf56dd0c77df7511b7cd2b2e3cdc8033f95f12136db020b7086eb270e"},"source":{"id":"2104.08821","kind":"arxiv","version":4},"verdict":{"created_at":"2026-05-15T07:44:10.339562Z","id":"9386c89c-7b0c-4e0d-9757-f550bca43508","model_set":{"reader":"grok-4.3"},"one_line_summary":"SimCSE achieves 76.3% unsupervised and 81.6% supervised Spearman's correlation on STS tasks with BERT-base, improving prior best results by 4.2% and 2.2% via simple contrastive learning.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Contrastive learning with standard dropout as the only noise produces sentence embeddings that match or beat prior supervised results.","strongest_claim":"our unsupervised and supervised models using BERT base achieve an average of 76.3% and 81.6% Spearman's correlation respectively, a 4.2% and 2.2% improvement compared to the previous best results. We also show -- both theoretically and empirically -- that the contrastive learning objective regularizes pre-trained embeddings' anisotropic space to be more uniform","weakest_assumption":"that standard dropout is sufficient as data augmentation to prevent representation collapse in the unsupervised contrastive objective, and that NLI entailment/contradiction pairs form appropriate positive and hard-negative pairs for learning general sentence embeddings."}},"verdict_id":"9386c89c-7b0c-4e0d-9757-f550bca43508"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:6ae4cdb441e89f658dc04fdb924445216af186245aae122455a46e0a84240cef","target":"record","created_at":"2026-05-17T23:38:53Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"0ba08acab8d5daef298eda40b969a5252fbad3b10cd5421f91a7fa217ca5a3b2","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-04-18T11:27:08Z","title_canon_sha256":"521eac38e6d56e5902f623741973ee63232a52df03edcbb497fbb9d6cc2c02e0"},"schema_version":"1.0","source":{"id":"2104.08821","kind":"arxiv","version":4}},"canonical_sha256":"48eb8cd584ca90ac3271ee3e606748130a9b5d5b1e43938a82e86154b8b68519","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"48eb8cd584ca90ac3271ee3e606748130a9b5d5b1e43938a82e86154b8b68519","first_computed_at":"2026-05-17T23:38:53.095688Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:53.095688Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"07T3FCm7zNBHOYdisHdxPUYmiobpHJ6WhkRZebKrpdF2prXV0yoyCE9ZyY/hHRNzUXG1cLihiGj7XTIFRE0iDQ==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:53.096185Z","signed_message":"canonical_sha256_bytes"},"source_id":"2104.08821","source_kind":"arxiv","source_version":4}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:6ae4cdb441e89f658dc04fdb924445216af186245aae122455a46e0a84240cef","sha256:cad7e423ee5d9e2994556c2ebf6261ea77e785c591e4af547c7bb0a01c13a668"],"state_sha256":"3d3e70fd7b11eac0b5b1f8394d2e2cb6ca5d92e4147089fb165b6f207d344eb0"}