{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CYEJQUMA4BWG3SRHBI45LBVPN6","short_pith_number":"pith:CYEJQUMA","schema_version":"1.0","canonical_sha256":"1608985180e06c6dca270a39d586af6f97f3d182c0c97f0d0c546c059d4e165b","source":{"kind":"arxiv","id":"2505.15239","version":1},"attestation_state":"computed","paper":{"title":"Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Christoph H. Lampert, Marco Mondelli, Peter S\\'uken\\'ik","submitted_at":"2025-05-21T08:16:03Z","abstract_excerpt":"The empirical emergence of neural collapse -- a surprising symmetry in the feature representations of the training data in the penultimate layer of deep neural networks -- has spurred a line of theoretical research aimed at its understanding. However, existing work focuses on data-agnostic models or, when data structure is taken into account, it remains limited to multi-layer perceptrons. Our paper fills both these gaps by analyzing modern architectures in a data-aware regime: we prove that global optima of deep regularized transformers and residual networks (ResNets) with LayerNorm trained wi"},"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":"2505.15239","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-05-21T08:16:03Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"41fa679f943c97b1b1ba845537ad4424df889cd6bc273cd8fa571edafbfb6608","abstract_canon_sha256":"34516a65b5039ec1ab6f8f30869c76d39a9d0cd50c45e8b86b3b99158d8aa1c1"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:06:42.573291Z","signature_b64":"MRaDrnBw5ygpOWN6rJBnSxRVGtskDFi6aIiKYF8pVnNjchl6Y6n4cS2Btlb8og9FXJoTzSD5iDZKmlAmx3FbDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1608985180e06c6dca270a39d586af6f97f3d182c0c97f0d0c546c059d4e165b","last_reissued_at":"2026-07-05T11:06:42.572828Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:06:42.572828Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Neural Collapse is Globally Optimal in Deep Regularized ResNets and Transformers","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Christoph H. Lampert, Marco Mondelli, Peter S\\'uken\\'ik","submitted_at":"2025-05-21T08:16:03Z","abstract_excerpt":"The empirical emergence of neural collapse -- a surprising symmetry in the feature representations of the training data in the penultimate layer of deep neural networks -- has spurred a line of theoretical research aimed at its understanding. However, existing work focuses on data-agnostic models or, when data structure is taken into account, it remains limited to multi-layer perceptrons. Our paper fills both these gaps by analyzing modern architectures in a data-aware regime: we prove that global optima of deep regularized transformers and residual networks (ResNets) with LayerNorm trained wi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.15239","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/2505.15239/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":"2505.15239","created_at":"2026-07-05T11:06:42.572881+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.15239v1","created_at":"2026-07-05T11:06:42.572881+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.15239","created_at":"2026-07-05T11:06:42.572881+00:00"},{"alias_kind":"pith_short_12","alias_value":"CYEJQUMA4BWG","created_at":"2026-07-05T11:06:42.572881+00:00"},{"alias_kind":"pith_short_16","alias_value":"CYEJQUMA4BWG3SRH","created_at":"2026-07-05T11:06:42.572881+00:00"},{"alias_kind":"pith_short_8","alias_value":"CYEJQUMA","created_at":"2026-07-05T11:06:42.572881+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26749","citing_title":"Structure Before Collapse: Transient semantic geometry in next-token prediction","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2510.01105","citing_title":"Geometric Analysis of Neural Regression Collapse via Intrinsic Dimension","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CYEJQUMA4BWG3SRHBI45LBVPN6","json":"https://pith.science/pith/CYEJQUMA4BWG3SRHBI45LBVPN6.json","graph_json":"https://pith.science/api/pith-number/CYEJQUMA4BWG3SRHBI45LBVPN6/graph.json","events_json":"https://pith.science/api/pith-number/CYEJQUMA4BWG3SRHBI45LBVPN6/events.json","paper":"https://pith.science/paper/CYEJQUMA"},"agent_actions":{"view_html":"https://pith.science/pith/CYEJQUMA4BWG3SRHBI45LBVPN6","download_json":"https://pith.science/pith/CYEJQUMA4BWG3SRHBI45LBVPN6.json","view_paper":"https://pith.science/paper/CYEJQUMA","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.15239&json=true","fetch_graph":"https://pith.science/api/pith-number/CYEJQUMA4BWG3SRHBI45LBVPN6/graph.json","fetch_events":"https://pith.science/api/pith-number/CYEJQUMA4BWG3SRHBI45LBVPN6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CYEJQUMA4BWG3SRHBI45LBVPN6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CYEJQUMA4BWG3SRHBI45LBVPN6/action/storage_attestation","attest_author":"https://pith.science/pith/CYEJQUMA4BWG3SRHBI45LBVPN6/action/author_attestation","sign_citation":"https://pith.science/pith/CYEJQUMA4BWG3SRHBI45LBVPN6/action/citation_signature","submit_replication":"https://pith.science/pith/CYEJQUMA4BWG3SRHBI45LBVPN6/action/replication_record"}},"created_at":"2026-07-05T11:06:42.572881+00:00","updated_at":"2026-07-05T11:06:42.572881+00:00"}