{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:KSSC3NVU26CYB6N6PFDUCMIMLT","short_pith_number":"pith:KSSC3NVU","schema_version":"1.0","canonical_sha256":"54a42db6b4d78580f9be794741310c5cc2e78416472cdde458660eb553222471","source":{"kind":"arxiv","id":"2410.08198","version":3},"attestation_state":"computed","paper":{"title":"Adam Exploits $\\ell_\\infty$-geometry of Loss Landscape via Coordinate-wise Adaptivity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Mohamad Amin Mohamadi, Shuo Xie, Zhiyuan Li","submitted_at":"2024-10-10T17:58:53Z","abstract_excerpt":"Adam outperforms SGD when training language models. Yet this advantage is not well-understood theoretically -- previous convergence analysis for Adam and SGD mainly focuses on the number of steps $T$ and is already minimax-optimal in non-convex cases, which are both $\\widetilde{O}(T^{-1/4})$. In this work, we argue that the exploitation of nice $\\ell_\\infty$-geometry is the key advantage of Adam over SGD. More specifically, we give a new convergence analysis for Adam under novel assumptions that loss is smooth under $\\ell_\\infty$-geometry rather than the more common $\\ell_2$-geometry, which yi"},"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":"2410.08198","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-10-10T17:58:53Z","cross_cats_sorted":[],"title_canon_sha256":"eaf0ebf5f357e671a4e59ade89415ddfef13243a18049ee42deb3efa1bb4b110","abstract_canon_sha256":"5439a593f75ed4001ecff6838226bf505165f79575e7e691dbf539f2d45f70a4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:49.677702Z","signature_b64":"fjN9MEEwCuIwxjSowCU+3RZa9SzMYZ82dRJIZau408Dq35E8iWb4Uxr6Y/P80J5KRyxHArQhI9hhJh8JEdUNCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"54a42db6b4d78580f9be794741310c5cc2e78416472cdde458660eb553222471","last_reissued_at":"2026-07-05T11:19:49.677231Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:49.677231Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Adam Exploits $\\ell_\\infty$-geometry of Loss Landscape via Coordinate-wise Adaptivity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Mohamad Amin Mohamadi, Shuo Xie, Zhiyuan Li","submitted_at":"2024-10-10T17:58:53Z","abstract_excerpt":"Adam outperforms SGD when training language models. Yet this advantage is not well-understood theoretically -- previous convergence analysis for Adam and SGD mainly focuses on the number of steps $T$ and is already minimax-optimal in non-convex cases, which are both $\\widetilde{O}(T^{-1/4})$. In this work, we argue that the exploitation of nice $\\ell_\\infty$-geometry is the key advantage of Adam over SGD. More specifically, we give a new convergence analysis for Adam under novel assumptions that loss is smooth under $\\ell_\\infty$-geometry rather than the more common $\\ell_2$-geometry, which yi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.08198","kind":"arxiv","version":3},"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/2410.08198/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":"2410.08198","created_at":"2026-07-05T11:19:49.677292+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.08198v3","created_at":"2026-07-05T11:19:49.677292+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.08198","created_at":"2026-07-05T11:19:49.677292+00:00"},{"alias_kind":"pith_short_12","alias_value":"KSSC3NVU26CY","created_at":"2026-07-05T11:19:49.677292+00:00"},{"alias_kind":"pith_short_16","alias_value":"KSSC3NVU26CYB6N6","created_at":"2026-07-05T11:19:49.677292+00:00"},{"alias_kind":"pith_short_8","alias_value":"KSSC3NVU","created_at":"2026-07-05T11:19:49.677292+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.11059","citing_title":"Uniform Scaling Limits in AdamW-Trained Transformers","ref_index":54,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KSSC3NVU26CYB6N6PFDUCMIMLT","json":"https://pith.science/pith/KSSC3NVU26CYB6N6PFDUCMIMLT.json","graph_json":"https://pith.science/api/pith-number/KSSC3NVU26CYB6N6PFDUCMIMLT/graph.json","events_json":"https://pith.science/api/pith-number/KSSC3NVU26CYB6N6PFDUCMIMLT/events.json","paper":"https://pith.science/paper/KSSC3NVU"},"agent_actions":{"view_html":"https://pith.science/pith/KSSC3NVU26CYB6N6PFDUCMIMLT","download_json":"https://pith.science/pith/KSSC3NVU26CYB6N6PFDUCMIMLT.json","view_paper":"https://pith.science/paper/KSSC3NVU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.08198&json=true","fetch_graph":"https://pith.science/api/pith-number/KSSC3NVU26CYB6N6PFDUCMIMLT/graph.json","fetch_events":"https://pith.science/api/pith-number/KSSC3NVU26CYB6N6PFDUCMIMLT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KSSC3NVU26CYB6N6PFDUCMIMLT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KSSC3NVU26CYB6N6PFDUCMIMLT/action/storage_attestation","attest_author":"https://pith.science/pith/KSSC3NVU26CYB6N6PFDUCMIMLT/action/author_attestation","sign_citation":"https://pith.science/pith/KSSC3NVU26CYB6N6PFDUCMIMLT/action/citation_signature","submit_replication":"https://pith.science/pith/KSSC3NVU26CYB6N6PFDUCMIMLT/action/replication_record"}},"created_at":"2026-07-05T11:19:49.677292+00:00","updated_at":"2026-07-05T11:19:49.677292+00:00"}