{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:RH672XDIC7QVDWOHB45MZBGHKP","short_pith_number":"pith:RH672XDI","schema_version":"1.0","canonical_sha256":"89fdfd5c6817e151d9c70f3acc84c753e54d4b0c193cf59c01871dad9ed0a940","source":{"kind":"arxiv","id":"2406.04443","version":3},"attestation_state":"computed","paper":{"title":"Clipping Improves Adam-Norm and AdaGrad-Norm when the Noise Is Heavy-Tailed","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC"],"primary_cat":"cs.LG","authors_text":"Aleksandr Beznosikov, Alexander Gasnikov, Andrei Semenov, Eduard Gorbunov, Martin Tak\\'a\\v{c}, Samuel Horv\\'ath, Savelii Chezhegov, Yaroslav Klyukin","submitted_at":"2024-06-06T18:49:10Z","abstract_excerpt":"Methods with adaptive stepsizes, such as AdaGrad and Adam, are essential for training modern Deep Learning models, especially Large Language Models. Typically, the noise in the stochastic gradients is heavy-tailed for the later ones. Gradient clipping provably helps to achieve good high-probability convergence for such noises. However, despite the similarity between AdaGrad/Adam and Clip-SGD, the current understanding of the high-probability convergence of AdaGrad/Adam-type methods is limited in this case. In this work, we prove that AdaGrad/Adam (and their delayed version) can have provably b"},"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":"2406.04443","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2024-06-06T18:49:10Z","cross_cats_sorted":["math.OC"],"title_canon_sha256":"abb18bb522547cae3408a07453c7c2221da059a73cf5421ee6e99226cbf56c65","abstract_canon_sha256":"36d15ce3e54aa32d8b46b901d551f2ccaf0341f8e179dd5a3e7b1c3be85aeb1c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:53:40.934181Z","signature_b64":"IswWaJ8bCj6QDQTOUDMDe8Gb61Cq+3zWjLv7YgFoT6DP06chu/JoYIpIE7c9WstSSHaHB3YRJFkkU62vT1C5DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"89fdfd5c6817e151d9c70f3acc84c753e54d4b0c193cf59c01871dad9ed0a940","last_reissued_at":"2026-07-05T11:53:40.933689Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:53:40.933689Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Clipping Improves Adam-Norm and AdaGrad-Norm when the Noise Is Heavy-Tailed","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC"],"primary_cat":"cs.LG","authors_text":"Aleksandr Beznosikov, Alexander Gasnikov, Andrei Semenov, Eduard Gorbunov, Martin Tak\\'a\\v{c}, Samuel Horv\\'ath, Savelii Chezhegov, Yaroslav Klyukin","submitted_at":"2024-06-06T18:49:10Z","abstract_excerpt":"Methods with adaptive stepsizes, such as AdaGrad and Adam, are essential for training modern Deep Learning models, especially Large Language Models. Typically, the noise in the stochastic gradients is heavy-tailed for the later ones. Gradient clipping provably helps to achieve good high-probability convergence for such noises. However, despite the similarity between AdaGrad/Adam and Clip-SGD, the current understanding of the high-probability convergence of AdaGrad/Adam-type methods is limited in this case. In this work, we prove that AdaGrad/Adam (and their delayed version) can have provably b"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2406.04443","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/2406.04443/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":"2406.04443","created_at":"2026-07-05T11:53:40.933746+00:00"},{"alias_kind":"arxiv_version","alias_value":"2406.04443v3","created_at":"2026-07-05T11:53:40.933746+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2406.04443","created_at":"2026-07-05T11:53:40.933746+00:00"},{"alias_kind":"pith_short_12","alias_value":"RH672XDIC7QV","created_at":"2026-07-05T11:53:40.933746+00:00"},{"alias_kind":"pith_short_16","alias_value":"RH672XDIC7QVDWOH","created_at":"2026-07-05T11:53:40.933746+00:00"},{"alias_kind":"pith_short_8","alias_value":"RH672XDI","created_at":"2026-07-05T11:53:40.933746+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.13287","citing_title":"Clipping Makes Distributed and Federated Asynchronous SGD Robust to Stragglers","ref_index":59,"is_internal_anchor":false},{"citing_arxiv_id":"2505.03736","citing_title":"Decentralized Nonconvex Optimization under Heavy-Tailed Noise: Normalization and Optimal Convergence","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22644","citing_title":"Why SGD is not Brownian Motion: A New Perspective on Stochastic Dynamics","ref_index":155,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15522","citing_title":"Stochastic Non-Smooth Convex Optimization with Unbounded Gradients","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11838","citing_title":"Gradient Clipping Beyond Vector Norms: A Spectral Approach for Matrix-Valued Parameters","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RH672XDIC7QVDWOHB45MZBGHKP","json":"https://pith.science/pith/RH672XDIC7QVDWOHB45MZBGHKP.json","graph_json":"https://pith.science/api/pith-number/RH672XDIC7QVDWOHB45MZBGHKP/graph.json","events_json":"https://pith.science/api/pith-number/RH672XDIC7QVDWOHB45MZBGHKP/events.json","paper":"https://pith.science/paper/RH672XDI"},"agent_actions":{"view_html":"https://pith.science/pith/RH672XDIC7QVDWOHB45MZBGHKP","download_json":"https://pith.science/pith/RH672XDIC7QVDWOHB45MZBGHKP.json","view_paper":"https://pith.science/paper/RH672XDI","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2406.04443&json=true","fetch_graph":"https://pith.science/api/pith-number/RH672XDIC7QVDWOHB45MZBGHKP/graph.json","fetch_events":"https://pith.science/api/pith-number/RH672XDIC7QVDWOHB45MZBGHKP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RH672XDIC7QVDWOHB45MZBGHKP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RH672XDIC7QVDWOHB45MZBGHKP/action/storage_attestation","attest_author":"https://pith.science/pith/RH672XDIC7QVDWOHB45MZBGHKP/action/author_attestation","sign_citation":"https://pith.science/pith/RH672XDIC7QVDWOHB45MZBGHKP/action/citation_signature","submit_replication":"https://pith.science/pith/RH672XDIC7QVDWOHB45MZBGHKP/action/replication_record"}},"created_at":"2026-07-05T11:53:40.933746+00:00","updated_at":"2026-07-05T11:53:40.933746+00:00"}