{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:4TNURL4F5O3ILHO4STELY76SYR","short_pith_number":"pith:4TNURL4F","schema_version":"1.0","canonical_sha256":"e4db48af85ebb6859ddc94c8bc7fd2c4737f0e6c3c04065a704a89bab0f4caeb","source":{"kind":"arxiv","id":"2303.15103","version":4},"attestation_state":"computed","paper":{"title":"Contrastive Learning Is Spectral Clustering On Similarity Graph","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Jingqin Yang, Yang Yuan, Yifan Zhang, Zhiquan Tan","submitted_at":"2023-03-27T11:13:35Z","abstract_excerpt":"Contrastive learning is a powerful self-supervised learning method, but we have a limited theoretical understanding of how it works and why it works. In this paper, we prove that contrastive learning with the standard InfoNCE loss is equivalent to spectral clustering on the similarity graph. Using this equivalence as the building block, we extend our analysis to the CLIP model and rigorously characterize how similar multi-modal objects are embedded together. Motivated by our theoretical insights, we introduce the Kernel-InfoNCE loss, incorporating mixtures of kernel functions that outperform t"},"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":"2303.15103","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-03-27T11:13:35Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"98f2e3534f5d9f21aa9e4e746c0e4279fc0275d7ef1bf73c207f42adc2ac04ca","abstract_canon_sha256":"075ce4abbd7e2df191ecf96bb82cc7124e509766203ba1ce197fb125534b81c1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:48:29.152443Z","signature_b64":"oaE9zc/IEpOrTtYcVfefU6afqZ7FJvsvzB4gIFpDd8yjUPL+Kbo3JLkG7dpF5Fglw/2WRn0STG5aJKtdSoOqDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e4db48af85ebb6859ddc94c8bc7fd2c4737f0e6c3c04065a704a89bab0f4caeb","last_reissued_at":"2026-07-05T07:48:29.151926Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:48:29.151926Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Contrastive Learning Is Spectral Clustering On Similarity Graph","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Jingqin Yang, Yang Yuan, Yifan Zhang, Zhiquan Tan","submitted_at":"2023-03-27T11:13:35Z","abstract_excerpt":"Contrastive learning is a powerful self-supervised learning method, but we have a limited theoretical understanding of how it works and why it works. In this paper, we prove that contrastive learning with the standard InfoNCE loss is equivalent to spectral clustering on the similarity graph. Using this equivalence as the building block, we extend our analysis to the CLIP model and rigorously characterize how similar multi-modal objects are embedded together. Motivated by our theoretical insights, we introduce the Kernel-InfoNCE loss, incorporating mixtures of kernel functions that outperform t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2303.15103","kind":"arxiv","version":4},"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/2303.15103/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":"2303.15103","created_at":"2026-07-05T07:48:29.151986+00:00"},{"alias_kind":"arxiv_version","alias_value":"2303.15103v4","created_at":"2026-07-05T07:48:29.151986+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2303.15103","created_at":"2026-07-05T07:48:29.151986+00:00"},{"alias_kind":"pith_short_12","alias_value":"4TNURL4F5O3I","created_at":"2026-07-05T07:48:29.151986+00:00"},{"alias_kind":"pith_short_16","alias_value":"4TNURL4F5O3ILHO4","created_at":"2026-07-05T07:48:29.151986+00:00"},{"alias_kind":"pith_short_8","alias_value":"4TNURL4F","created_at":"2026-07-05T07:48:29.151986+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22873","citing_title":"SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning","ref_index":85,"is_internal_anchor":false},{"citing_arxiv_id":"2606.22873","citing_title":"SingGuard: A Policy-Adaptive Multimodal LLM Guardrail with Dynamic Reasoning","ref_index":85,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4TNURL4F5O3ILHO4STELY76SYR","json":"https://pith.science/pith/4TNURL4F5O3ILHO4STELY76SYR.json","graph_json":"https://pith.science/api/pith-number/4TNURL4F5O3ILHO4STELY76SYR/graph.json","events_json":"https://pith.science/api/pith-number/4TNURL4F5O3ILHO4STELY76SYR/events.json","paper":"https://pith.science/paper/4TNURL4F"},"agent_actions":{"view_html":"https://pith.science/pith/4TNURL4F5O3ILHO4STELY76SYR","download_json":"https://pith.science/pith/4TNURL4F5O3ILHO4STELY76SYR.json","view_paper":"https://pith.science/paper/4TNURL4F","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2303.15103&json=true","fetch_graph":"https://pith.science/api/pith-number/4TNURL4F5O3ILHO4STELY76SYR/graph.json","fetch_events":"https://pith.science/api/pith-number/4TNURL4F5O3ILHO4STELY76SYR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4TNURL4F5O3ILHO4STELY76SYR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4TNURL4F5O3ILHO4STELY76SYR/action/storage_attestation","attest_author":"https://pith.science/pith/4TNURL4F5O3ILHO4STELY76SYR/action/author_attestation","sign_citation":"https://pith.science/pith/4TNURL4F5O3ILHO4STELY76SYR/action/citation_signature","submit_replication":"https://pith.science/pith/4TNURL4F5O3ILHO4STELY76SYR/action/replication_record"}},"created_at":"2026-07-05T07:48:29.151986+00:00","updated_at":"2026-07-05T07:48:29.151986+00:00"}