{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:TJHARD3F6XMS45SE7CLM2PH3UB","short_pith_number":"pith:TJHARD3F","schema_version":"1.0","canonical_sha256":"9a4e088f65f5d92e7644f896cd3cfba07796aad95c7048809e0552094da8b0ee","source":{"kind":"arxiv","id":"2509.03378","version":10},"attestation_state":"computed","paper":{"title":"Understanding and Improving Shampoo and SOAP via Kullback-Leibler Minimization","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Felix Dangel, Roger B. Grosse, Runa Eschenhagen, Scott C. Lowe, Wu Lin, Zikun Xu","submitted_at":"2025-09-03T14:55:15Z","abstract_excerpt":"Shampoo and its efficient variant, SOAP, employ structured second-moment estimations and have shown strong performance for training neural networks (NNs). In practice, however, Shampoo typically requires step-size grafting with Adam to be competitive, and SOAP mitigates this by applying Adam in Shampoo's eigenbasis -- at the cost of additional memory overhead from Adam in both methods. Prior analyses have largely relied on the Frobenius norm to motivate these estimation schemes. We instead recast their estimation procedures as covariance estimation under Kullback-Leibler (KL) divergence minimi"},"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":"2509.03378","kind":"arxiv","version":10},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"stat.ML","submitted_at":"2025-09-03T14:55:15Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"9c5f1023d0e5dd4280c56819414b34d08b8b76d954a5f6f1637c07f8f32c926a","abstract_canon_sha256":"afc82a822ceaae3820b9f0aec38c6fe2d19dc9bfdd97f5d185433fb64fd16456"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T03:13:47.266907Z","signature_b64":"OKNmyH2ze3zhE/dDWlt84v6iU8HT07HJ2wu9I4rVZi263zxWY7ODx5zsRqvP/GxPtYM6plFO1GzlfI2WQPaUBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9a4e088f65f5d92e7644f896cd3cfba07796aad95c7048809e0552094da8b0ee","last_reissued_at":"2026-06-23T03:13:47.266410Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T03:13:47.266410Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding and Improving Shampoo and SOAP via Kullback-Leibler Minimization","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Felix Dangel, Roger B. Grosse, Runa Eschenhagen, Scott C. Lowe, Wu Lin, Zikun Xu","submitted_at":"2025-09-03T14:55:15Z","abstract_excerpt":"Shampoo and its efficient variant, SOAP, employ structured second-moment estimations and have shown strong performance for training neural networks (NNs). In practice, however, Shampoo typically requires step-size grafting with Adam to be competitive, and SOAP mitigates this by applying Adam in Shampoo's eigenbasis -- at the cost of additional memory overhead from Adam in both methods. Prior analyses have largely relied on the Frobenius norm to motivate these estimation schemes. We instead recast their estimation procedures as covariance estimation under Kullback-Leibler (KL) divergence minimi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.03378","kind":"arxiv","version":10},"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/2509.03378/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":"2509.03378","created_at":"2026-06-23T03:13:47.266472+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.03378v10","created_at":"2026-06-23T03:13:47.266472+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.03378","created_at":"2026-06-23T03:13:47.266472+00:00"},{"alias_kind":"pith_short_12","alias_value":"TJHARD3F6XMS","created_at":"2026-06-23T03:13:47.266472+00:00"},{"alias_kind":"pith_short_16","alias_value":"TJHARD3F6XMS45SE","created_at":"2026-06-23T03:13:47.266472+00:00"},{"alias_kind":"pith_short_8","alias_value":"TJHARD3F","created_at":"2026-06-23T03:13:47.266472+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":7,"sample":[{"citing_arxiv_id":"2606.06418","citing_title":"Double Preconditioning (DoPr): Optimization for Test-Time Performance, not Validation Loss","ref_index":170,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18106","citing_title":"Symmetry-Compatible Principle for Optimizer Design: Embeddings, LM Heads, SwiGLU MLPs, and MoE Routers","ref_index":106,"is_internal_anchor":true},{"citing_arxiv_id":"2510.10777","citing_title":"Preconditioned Norms: A Unified Framework for Steepest Descent, Quasi-Newton and Adaptive Methods","ref_index":13,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18106","citing_title":"Symmetry-Compatible Principle for Optimizer Design: Embeddings, LM Heads, SwiGLU MLPs, and MoE Routers","ref_index":104,"is_internal_anchor":true},{"citing_arxiv_id":"2605.16184","citing_title":"Runtime-Orchestrated Second-Order Optimization for Scalable LLM Training","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06316","citing_title":"Pro-KLShampoo: Projected KL-Shampoo with Whitening Recovered by Orthogonalization","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2604.05230","citing_title":"Curvature-Aware Optimization for High-Accuracy Physics-Informed Neural Networks","ref_index":50,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/TJHARD3F6XMS45SE7CLM2PH3UB","json":"https://pith.science/pith/TJHARD3F6XMS45SE7CLM2PH3UB.json","graph_json":"https://pith.science/api/pith-number/TJHARD3F6XMS45SE7CLM2PH3UB/graph.json","events_json":"https://pith.science/api/pith-number/TJHARD3F6XMS45SE7CLM2PH3UB/events.json","paper":"https://pith.science/paper/TJHARD3F"},"agent_actions":{"view_html":"https://pith.science/pith/TJHARD3F6XMS45SE7CLM2PH3UB","download_json":"https://pith.science/pith/TJHARD3F6XMS45SE7CLM2PH3UB.json","view_paper":"https://pith.science/paper/TJHARD3F","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.03378&json=true","fetch_graph":"https://pith.science/api/pith-number/TJHARD3F6XMS45SE7CLM2PH3UB/graph.json","fetch_events":"https://pith.science/api/pith-number/TJHARD3F6XMS45SE7CLM2PH3UB/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/TJHARD3F6XMS45SE7CLM2PH3UB/action/timestamp_anchor","attest_storage":"https://pith.science/pith/TJHARD3F6XMS45SE7CLM2PH3UB/action/storage_attestation","attest_author":"https://pith.science/pith/TJHARD3F6XMS45SE7CLM2PH3UB/action/author_attestation","sign_citation":"https://pith.science/pith/TJHARD3F6XMS45SE7CLM2PH3UB/action/citation_signature","submit_replication":"https://pith.science/pith/TJHARD3F6XMS45SE7CLM2PH3UB/action/replication_record"}},"created_at":"2026-06-23T03:13:47.266472+00:00","updated_at":"2026-06-23T03:13:47.266472+00:00"}