{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:IYHWGC3FVITNJHE2HKKPOXTFX2","short_pith_number":"pith:IYHWGC3F","schema_version":"1.0","canonical_sha256":"460f630b65aa26d49c9a3a94f75e65bebbd95699c8da782d92748a3c0e6658d6","source":{"kind":"arxiv","id":"2601.07326","version":3},"attestation_state":"computed","paper":{"title":"Convergence Rate Analysis of the AdamW-Style Shampoo: Unifying One-Sided and Two-Sided Preconditioning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"AdamW-style Shampoo unifies one-sided and two-sided preconditioning to achieve an SGD-like convergence rate in nuclear norm.","cross_cats":["cs.LG"],"primary_cat":"math.OC","authors_text":"Huan Li, Yiming Dong, Zhouchen Lin","submitted_at":"2026-01-12T08:51:03Z","abstract_excerpt":"This paper studies the AdamW-style Shampoo optimizer, an effective implementation of classical Shampoo that notably won the external tuning track of the AlgoPerf neural network training algorithm competition. Our analysis unifies one-sided and two-sided preconditioning and establishes the convergence rate $\\frac{1}{K}\\sum_{k=1}^K E\\left[\\|\\nabla f(X_k)\\|_*\\right]\\leq O(\\frac{\\sqrt{m+n}C}{K^{1/4}})$ measured by nuclear norm, where $K$ represents the iteration number, $(m,n)$ denotes the size of matrix parameters, and $C$ matches the constant in the optimal convergence rate of SGD. Theoretically"},"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":true},"canonical_record":{"source":{"id":"2601.07326","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"math.OC","submitted_at":"2026-01-12T08:51:03Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"69c3092ca2f2c829965d7a48a0aa50ef530a4f53d345f7b0a9bedd43b5260bb4","abstract_canon_sha256":"acf196fa014f2c358fe7b80ca3cdbbfc37e6d17a27e0372a5eb3f25ab39b88c6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:02.947633Z","signature_b64":"bWxfp5o0HKHlb3dNAIKYBnY85Ipmo7TbLHTt/GtGV+YgrmNKLCqVsAxgTzKFxJ04fKwjyopagb751T3cCnhYAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"460f630b65aa26d49c9a3a94f75e65bebbd95699c8da782d92748a3c0e6658d6","last_reissued_at":"2026-05-26T02:04:02.946717Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:02.946717Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Convergence Rate Analysis of the AdamW-Style Shampoo: Unifying One-Sided and Two-Sided Preconditioning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"AdamW-style Shampoo unifies one-sided and two-sided preconditioning to achieve an SGD-like convergence rate in nuclear norm.","cross_cats":["cs.LG"],"primary_cat":"math.OC","authors_text":"Huan Li, Yiming Dong, Zhouchen Lin","submitted_at":"2026-01-12T08:51:03Z","abstract_excerpt":"This paper studies the AdamW-style Shampoo optimizer, an effective implementation of classical Shampoo that notably won the external tuning track of the AlgoPerf neural network training algorithm competition. Our analysis unifies one-sided and two-sided preconditioning and establishes the convergence rate $\\frac{1}{K}\\sum_{k=1}^K E\\left[\\|\\nabla f(X_k)\\|_*\\right]\\leq O(\\frac{\\sqrt{m+n}C}{K^{1/4}})$ measured by nuclear norm, where $K$ represents the iteration number, $(m,n)$ denotes the size of matrix parameters, and $C$ matches the constant in the optimal convergence rate of SGD. Theoretically"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our analysis unifies one-sided and two-sided preconditioning and establishes the convergence rate 1/K sum_{k=1}^K E[||nabla f(X_k)||_*] <= O(sqrt(m+n) C / K^{1/4}) measured by nuclear norm.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The derivation relies on standard stochastic-optimization assumptions (smoothness, bounded variance or similar) together with the specific form of the AdamW-style preconditioner; the abstract does not list the precise conditions under which the nuclear-norm bound holds.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"AdamW-style Shampoo unifies one- and two-sided preconditioning and achieves convergence rate O(sqrt(m+n) C / K^{1/4}) in nuclear norm, analogous to SGD's optimal rate when nuclear norm scales as Theta(sqrt(m+n)) times Frobenius norm.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"AdamW-style Shampoo unifies one-sided and two-sided preconditioning to achieve an SGD-like convergence rate in nuclear norm.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"a4bdbdb09ed1cffa0deb2ab91dc0a7f392349d920126dfc64954e4f5c4e71b65"},"source":{"id":"2601.07326","kind":"arxiv","version":3},"verdict":{"id":"d8814ebf-967b-453e-86ad-1bf2f7e16741","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T15:33:28.762115Z","strongest_claim":"Our analysis unifies one-sided and two-sided preconditioning and establishes the convergence rate 1/K sum_{k=1}^K E[||nabla f(X_k)||_*] <= O(sqrt(m+n) C / K^{1/4}) measured by nuclear norm.","one_line_summary":"AdamW-style Shampoo unifies one- and two-sided preconditioning and achieves convergence rate O(sqrt(m+n) C / K^{1/4}) in nuclear norm, analogous to SGD's optimal rate when nuclear norm scales as Theta(sqrt(m+n)) times Frobenius norm.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The derivation relies on standard stochastic-optimization assumptions (smoothness, bounded variance or similar) together with the specific form of the AdamW-style preconditioner; the abstract does not list the precise conditions under which the nuclear-norm bound holds.","pith_extraction_headline":"AdamW-style Shampoo unifies one-sided and two-sided preconditioning to achieve an SGD-like convergence rate in nuclear norm."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2601.07326/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":2,"snapshot_sha256":"87a337bb60f7c45b5faa450549507a141a5309e4028a2021a67d78451d971949"},"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":"2601.07326","created_at":"2026-05-26T02:04:02.946816+00:00"},{"alias_kind":"arxiv_version","alias_value":"2601.07326v3","created_at":"2026-05-26T02:04:02.946816+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2601.07326","created_at":"2026-05-26T02:04:02.946816+00:00"},{"alias_kind":"pith_short_12","alias_value":"IYHWGC3FVITN","created_at":"2026-05-26T02:04:02.946816+00:00"},{"alias_kind":"pith_short_16","alias_value":"IYHWGC3FVITNJHE2","created_at":"2026-05-26T02:04:02.946816+00:00"},{"alias_kind":"pith_short_8","alias_value":"IYHWGC3F","created_at":"2026-05-26T02:04:02.946816+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.18106","citing_title":"Symmetry-Compatible Principle for Optimizer Design: Embeddings, LM Heads, SwiGLU MLPs, and MoE Routers","ref_index":96,"is_internal_anchor":true},{"citing_arxiv_id":"2604.21616","citing_title":"Convergence Rate Analysis of SOAP with Arbitrary Orthogonal Projection Matrices","ref_index":22,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":2,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IYHWGC3FVITNJHE2HKKPOXTFX2","json":"https://pith.science/pith/IYHWGC3FVITNJHE2HKKPOXTFX2.json","graph_json":"https://pith.science/api/pith-number/IYHWGC3FVITNJHE2HKKPOXTFX2/graph.json","events_json":"https://pith.science/api/pith-number/IYHWGC3FVITNJHE2HKKPOXTFX2/events.json","paper":"https://pith.science/paper/IYHWGC3F"},"agent_actions":{"view_html":"https://pith.science/pith/IYHWGC3FVITNJHE2HKKPOXTFX2","download_json":"https://pith.science/pith/IYHWGC3FVITNJHE2HKKPOXTFX2.json","view_paper":"https://pith.science/paper/IYHWGC3F","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2601.07326&json=true","fetch_graph":"https://pith.science/api/pith-number/IYHWGC3FVITNJHE2HKKPOXTFX2/graph.json","fetch_events":"https://pith.science/api/pith-number/IYHWGC3FVITNJHE2HKKPOXTFX2/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IYHWGC3FVITNJHE2HKKPOXTFX2/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IYHWGC3FVITNJHE2HKKPOXTFX2/action/storage_attestation","attest_author":"https://pith.science/pith/IYHWGC3FVITNJHE2HKKPOXTFX2/action/author_attestation","sign_citation":"https://pith.science/pith/IYHWGC3FVITNJHE2HKKPOXTFX2/action/citation_signature","submit_replication":"https://pith.science/pith/IYHWGC3FVITNJHE2HKKPOXTFX2/action/replication_record"}},"created_at":"2026-05-26T02:04:02.946816+00:00","updated_at":"2026-05-26T02:04:02.946816+00:00"}