{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:HBT4R4PSK4QKPOR3RAJDJSFV55","short_pith_number":"pith:HBT4R4PS","schema_version":"1.0","canonical_sha256":"3867c8f1f25720a7ba3b881234c8b5ef4c7ddf79c94d820d3586a3eb0fd9d91e","source":{"kind":"arxiv","id":"2107.01152","version":1},"attestation_state":"computed","paper":{"title":"Simpler, Faster, Stronger: Breaking The log-K Curse On Contrastive Learners With FlatNCE","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.IT","cs.LG","math.IT"],"primary_cat":"stat.ML","authors_text":"Belinda Zeng, Chenyang Tao, Fan Li, Junya Chen, Lawrence Carin, Liqun Chen, Qing Guo, Shuyang Gao, Tagyoung Chung, Wenlian Lu, Xuan Li, Yi Xu, Zhe Gan","submitted_at":"2021-07-02T15:50:43Z","abstract_excerpt":"InfoNCE-based contrastive representation learners, such as SimCLR, have been tremendously successful in recent years. However, these contrastive schemes are notoriously resource demanding, as their effectiveness breaks down with small-batch training (i.e., the log-K curse, whereas K is the batch-size). In this work, we reveal mathematically why contrastive learners fail in the small-batch-size regime, and present a novel simple, non-trivial contrastive objective named FlatNCE, which fixes this issue. Unlike InfoNCE, our FlatNCE no longer explicitly appeals to a discriminative classification go"},"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":"2107.01152","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2021-07-02T15:50:43Z","cross_cats_sorted":["cs.AI","cs.CV","cs.IT","cs.LG","math.IT"],"title_canon_sha256":"6fc9b39a221a0e1b569fb2af7090add1a59aaf57c864d54bdd42c74ecb22b2a4","abstract_canon_sha256":"273d3407459c72cf5318d692a5158860e8ea1dc4099303aaab547ddb5d01d55e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:54:37.650461Z","signature_b64":"Y4Sw6KITDXw5dVmt87DMSpMbG07HRBl+2uWHFxEGAr3MEevh7Y54OEoPz10bbnP8P3+9Cdg+HI0ZVqpP+9caDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3867c8f1f25720a7ba3b881234c8b5ef4c7ddf79c94d820d3586a3eb0fd9d91e","last_reissued_at":"2026-07-05T02:54:37.650038Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:54:37.650038Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Simpler, Faster, Stronger: Breaking The log-K Curse On Contrastive Learners With FlatNCE","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.CV","cs.IT","cs.LG","math.IT"],"primary_cat":"stat.ML","authors_text":"Belinda Zeng, Chenyang Tao, Fan Li, Junya Chen, Lawrence Carin, Liqun Chen, Qing Guo, Shuyang Gao, Tagyoung Chung, Wenlian Lu, Xuan Li, Yi Xu, Zhe Gan","submitted_at":"2021-07-02T15:50:43Z","abstract_excerpt":"InfoNCE-based contrastive representation learners, such as SimCLR, have been tremendously successful in recent years. However, these contrastive schemes are notoriously resource demanding, as their effectiveness breaks down with small-batch training (i.e., the log-K curse, whereas K is the batch-size). In this work, we reveal mathematically why contrastive learners fail in the small-batch-size regime, and present a novel simple, non-trivial contrastive objective named FlatNCE, which fixes this issue. Unlike InfoNCE, our FlatNCE no longer explicitly appeals to a discriminative classification go"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2107.01152","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/2107.01152/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":"2107.01152","created_at":"2026-07-05T02:54:37.650094+00:00"},{"alias_kind":"arxiv_version","alias_value":"2107.01152v1","created_at":"2026-07-05T02:54:37.650094+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2107.01152","created_at":"2026-07-05T02:54:37.650094+00:00"},{"alias_kind":"pith_short_12","alias_value":"HBT4R4PSK4QK","created_at":"2026-07-05T02:54:37.650094+00:00"},{"alias_kind":"pith_short_16","alias_value":"HBT4R4PSK4QKPOR3","created_at":"2026-07-05T02:54:37.650094+00:00"},{"alias_kind":"pith_short_8","alias_value":"HBT4R4PS","created_at":"2026-07-05T02:54:37.650094+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.07211","citing_title":"HARMONY: Bridging the Personalization-Generalization Gap by Mitigating Representation Skew in Heterogeneous Split Federated Learning","ref_index":2,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HBT4R4PSK4QKPOR3RAJDJSFV55","json":"https://pith.science/pith/HBT4R4PSK4QKPOR3RAJDJSFV55.json","graph_json":"https://pith.science/api/pith-number/HBT4R4PSK4QKPOR3RAJDJSFV55/graph.json","events_json":"https://pith.science/api/pith-number/HBT4R4PSK4QKPOR3RAJDJSFV55/events.json","paper":"https://pith.science/paper/HBT4R4PS"},"agent_actions":{"view_html":"https://pith.science/pith/HBT4R4PSK4QKPOR3RAJDJSFV55","download_json":"https://pith.science/pith/HBT4R4PSK4QKPOR3RAJDJSFV55.json","view_paper":"https://pith.science/paper/HBT4R4PS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2107.01152&json=true","fetch_graph":"https://pith.science/api/pith-number/HBT4R4PSK4QKPOR3RAJDJSFV55/graph.json","fetch_events":"https://pith.science/api/pith-number/HBT4R4PSK4QKPOR3RAJDJSFV55/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HBT4R4PSK4QKPOR3RAJDJSFV55/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HBT4R4PSK4QKPOR3RAJDJSFV55/action/storage_attestation","attest_author":"https://pith.science/pith/HBT4R4PSK4QKPOR3RAJDJSFV55/action/author_attestation","sign_citation":"https://pith.science/pith/HBT4R4PSK4QKPOR3RAJDJSFV55/action/citation_signature","submit_replication":"https://pith.science/pith/HBT4R4PSK4QKPOR3RAJDJSFV55/action/replication_record"}},"created_at":"2026-07-05T02:54:37.650094+00:00","updated_at":"2026-07-05T02:54:37.650094+00:00"}