{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:HIZMU5ULBLZB5JVSPSTS4ODD7A","short_pith_number":"pith:HIZMU5UL","schema_version":"1.0","canonical_sha256":"3a32ca768b0af21ea6b27ca72e3863f839133dc8478f30ea8e1b34b7eb39b188","source":{"kind":"arxiv","id":"2605.13943","version":1},"attestation_state":"computed","paper":{"title":"A Unified Geometric Framework for Weighted Contrastive Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Weighted InfoNCE objectives correspond to distance geometry problems whose solutions fix the geometry of optimal embeddings.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Benoit Dufumier, Edouard Duchesnay, Raphael Vock","submitted_at":"2026-05-13T17:48:36Z","abstract_excerpt":"Contrastive learning (CL) aims to preserve relational structure between samples by learning representations that reflect a similarity graph. Yet, the geometry of the resulting embeddings remains poorly understood. Here we show that weighted InfoNCE objectives can be interpreted as Distance Geometry Problems, where the weighting scheme specifies the target geometry to be realized by the representation. This viewpoint yields exact characterizations of the optimal embeddings for several supervised and weakly supervised objectives. In supervised classification, both SupCon and Soft SupCon (a dense"},"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":true,"formal_links_present":true},"canonical_record":{"source":{"id":"2605.13943","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T17:48:36Z","cross_cats_sorted":[],"title_canon_sha256":"4b3ccdb39002a04cbbba6ea41f1b392a68938f58b3b1981e9a5acdd5f2bbfe5b","abstract_canon_sha256":"a499e9897a23916596d8abf70f52558e333a6cc92b72e0c7cbbe71dea9cb09c5"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:13.831723Z","signature_b64":"/d9/4WLtF979KeUoO1tXZl3m6Pw2yT07Itt6gd+3RQmMGRPJbibOka+83XwduWvNoAOqqQzO7dbYItuDVEYrCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3a32ca768b0af21ea6b27ca72e3863f839133dc8478f30ea8e1b34b7eb39b188","last_reissued_at":"2026-05-17T23:39:13.831100Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:13.831100Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Unified Geometric Framework for Weighted Contrastive Learning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Weighted InfoNCE objectives correspond to distance geometry problems whose solutions fix the geometry of optimal embeddings.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Benoit Dufumier, Edouard Duchesnay, Raphael Vock","submitted_at":"2026-05-13T17:48:36Z","abstract_excerpt":"Contrastive learning (CL) aims to preserve relational structure between samples by learning representations that reflect a similarity graph. Yet, the geometry of the resulting embeddings remains poorly understood. Here we show that weighted InfoNCE objectives can be interpreted as Distance Geometry Problems, where the weighting scheme specifies the target geometry to be realized by the representation. This viewpoint yields exact characterizations of the optimal embeddings for several supervised and weakly supervised objectives. In supervised classification, both SupCon and Soft SupCon (a dense"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"weighted InfoNCE objectives can be interpreted as Distance Geometry Problems, where the weighting scheme specifies the target geometry to be realized by the representation. This viewpoint yields exact characterizations of the optimal embeddings for several supervised and weakly supervised objectives.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the contrastive objective can be exactly recast as a distance geometry problem whose solution is attained by gradient descent on the embedding parameters, without additional constraints from finite batch sizes, temperature scaling, or optimization dynamics.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Weighted InfoNCE objectives realize specific target geometries in embedding space, with SupCon producing size-dependent inter-class similarities under imbalance while Soft SupCon and certain continuous variants preserve regular simplex or unique optima.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Weighted InfoNCE objectives correspond to distance geometry problems whose solutions fix the geometry of optimal embeddings.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"09bed2f1bcb7cb736aeeec42bc370cd01cc3eb8af1e275f5bf9c5234d6dda61a"},"source":{"id":"2605.13943","kind":"arxiv","version":1},"verdict":{"id":"a2b2490b-aaaa-4a85-a467-8044d931e763","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T06:09:35.431923Z","strongest_claim":"weighted InfoNCE objectives can be interpreted as Distance Geometry Problems, where the weighting scheme specifies the target geometry to be realized by the representation. This viewpoint yields exact characterizations of the optimal embeddings for several supervised and weakly supervised objectives.","one_line_summary":"Weighted InfoNCE objectives realize specific target geometries in embedding space, with SupCon producing size-dependent inter-class similarities under imbalance while Soft SupCon and certain continuous variants preserve regular simplex or unique optima.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the contrastive objective can be exactly recast as a distance geometry problem whose solution is attained by gradient descent on the embedding parameters, without additional constraints from finite batch sizes, temperature scaling, or optimization dynamics.","pith_extraction_headline":"Weighted InfoNCE objectives correspond to distance geometry problems whose solutions fix the geometry of optimal embeddings."},"references":{"count":51,"sample":[{"doi":"","year":2021,"title":"International Conference on Machine Learning , pages=","work_id":"24cd7fea-9b0e-401f-8035-63bbd69e1da2","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1981,"title":"Journal of Mathematical Psychology , volume=","work_id":"69e26ed3-4ce3-4c66-9211-9baf9fdba474","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1975,"title":"Cognitive psychology , volume=","work_id":"d5e3fae0-0758-4d2c-b0be-60d280df3b9c","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Advances in Neural Information Processing Systems , volume=","work_id":"7fd011c4-a13a-46b3-899c-e9a0da6014ec","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"arXiv preprint arXiv:2407.18134 , year=","work_id":"67f2665e-3d0f-4cb5-880b-873decc3c8bc","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":51,"snapshot_sha256":"fbfe34b6fc1dd6b2df40dc5d5540ad5c4df66cf1ef8ffd77e1b423794e738431","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"03d683225b098407b0cb54dc3652c01736651c8b877b61b112bd64a702af02d6"},"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":"2605.13943","created_at":"2026-05-17T23:39:13.831181+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.13943v1","created_at":"2026-05-17T23:39:13.831181+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13943","created_at":"2026-05-17T23:39:13.831181+00:00"},{"alias_kind":"pith_short_12","alias_value":"HIZMU5ULBLZB","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"HIZMU5ULBLZB5JVS","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"HIZMU5UL","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HIZMU5ULBLZB5JVSPSTS4ODD7A","json":"https://pith.science/pith/HIZMU5ULBLZB5JVSPSTS4ODD7A.json","graph_json":"https://pith.science/api/pith-number/HIZMU5ULBLZB5JVSPSTS4ODD7A/graph.json","events_json":"https://pith.science/api/pith-number/HIZMU5ULBLZB5JVSPSTS4ODD7A/events.json","paper":"https://pith.science/paper/HIZMU5UL"},"agent_actions":{"view_html":"https://pith.science/pith/HIZMU5ULBLZB5JVSPSTS4ODD7A","download_json":"https://pith.science/pith/HIZMU5ULBLZB5JVSPSTS4ODD7A.json","view_paper":"https://pith.science/paper/HIZMU5UL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.13943&json=true","fetch_graph":"https://pith.science/api/pith-number/HIZMU5ULBLZB5JVSPSTS4ODD7A/graph.json","fetch_events":"https://pith.science/api/pith-number/HIZMU5ULBLZB5JVSPSTS4ODD7A/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HIZMU5ULBLZB5JVSPSTS4ODD7A/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HIZMU5ULBLZB5JVSPSTS4ODD7A/action/storage_attestation","attest_author":"https://pith.science/pith/HIZMU5ULBLZB5JVSPSTS4ODD7A/action/author_attestation","sign_citation":"https://pith.science/pith/HIZMU5ULBLZB5JVSPSTS4ODD7A/action/citation_signature","submit_replication":"https://pith.science/pith/HIZMU5ULBLZB5JVSPSTS4ODD7A/action/replication_record"}},"created_at":"2026-05-17T23:39:13.831181+00:00","updated_at":"2026-05-17T23:39:13.831181+00:00"}