{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:DE527A2RJAHUUPRXJVQO4B5M7W","short_pith_number":"pith:DE527A2R","schema_version":"1.0","canonical_sha256":"193baf8351480f4a3e374d60ee07acfda595956004fa8aebf936b7a8f0107ee2","source":{"kind":"arxiv","id":"2302.02971","version":1},"attestation_state":"computed","paper":{"title":"U-Clip: On-Average Unbiased Stochastic Gradient Clipping","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Bryn Elesedy, Marcus Hutter","submitted_at":"2023-02-06T18:01:38Z","abstract_excerpt":"U-Clip is a simple amendment to gradient clipping that can be applied to any iterative gradient optimization algorithm. Like regular clipping, U-Clip involves using gradients that are clipped to a prescribed size (e.g. with component wise or norm based clipping) but instead of discarding the clipped portion of the gradient, U-Clip maintains a buffer of these values that is added to the gradients on the next iteration (before clipping). We show that the cumulative bias of the U-Clip updates is bounded by a constant. This implies that the clipped updates are unbiased on average. Convergence foll"},"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":"2302.02971","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-02-06T18:01:38Z","cross_cats_sorted":["math.OC","stat.ML"],"title_canon_sha256":"b36bf03c73a2d42f714118da00144280ef8231238f4d47ec40a5c29747d3720f","abstract_canon_sha256":"03705537157ca24c88f8b7a7158363c336526065e10c090c2b39e3118fe1ba5f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:39:08.731477Z","signature_b64":"0SvDpSL0M782+6s1aXKhlpT7K4u9+sgDqpHn3iXBDMegcjce0PHRVS9q6IlFeCIhzPgJgqmlmAR1EubX+V8mCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"193baf8351480f4a3e374d60ee07acfda595956004fa8aebf936b7a8f0107ee2","last_reissued_at":"2026-07-05T05:39:08.731071Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:39:08.731071Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"U-Clip: On-Average Unbiased Stochastic Gradient Clipping","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Bryn Elesedy, Marcus Hutter","submitted_at":"2023-02-06T18:01:38Z","abstract_excerpt":"U-Clip is a simple amendment to gradient clipping that can be applied to any iterative gradient optimization algorithm. Like regular clipping, U-Clip involves using gradients that are clipped to a prescribed size (e.g. with component wise or norm based clipping) but instead of discarding the clipped portion of the gradient, U-Clip maintains a buffer of these values that is added to the gradients on the next iteration (before clipping). We show that the cumulative bias of the U-Clip updates is bounded by a constant. This implies that the clipped updates are unbiased on average. Convergence foll"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2302.02971","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/2302.02971/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":"2302.02971","created_at":"2026-07-05T05:39:08.731126+00:00"},{"alias_kind":"arxiv_version","alias_value":"2302.02971v1","created_at":"2026-07-05T05:39:08.731126+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2302.02971","created_at":"2026-07-05T05:39:08.731126+00:00"},{"alias_kind":"pith_short_12","alias_value":"DE527A2RJAHU","created_at":"2026-07-05T05:39:08.731126+00:00"},{"alias_kind":"pith_short_16","alias_value":"DE527A2RJAHUUPRX","created_at":"2026-07-05T05:39:08.731126+00:00"},{"alias_kind":"pith_short_8","alias_value":"DE527A2R","created_at":"2026-07-05T05:39:08.731126+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2501.00265","citing_title":"Outlier-Robust Training of Machine Learning Models","ref_index":32,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DE527A2RJAHUUPRXJVQO4B5M7W","json":"https://pith.science/pith/DE527A2RJAHUUPRXJVQO4B5M7W.json","graph_json":"https://pith.science/api/pith-number/DE527A2RJAHUUPRXJVQO4B5M7W/graph.json","events_json":"https://pith.science/api/pith-number/DE527A2RJAHUUPRXJVQO4B5M7W/events.json","paper":"https://pith.science/paper/DE527A2R"},"agent_actions":{"view_html":"https://pith.science/pith/DE527A2RJAHUUPRXJVQO4B5M7W","download_json":"https://pith.science/pith/DE527A2RJAHUUPRXJVQO4B5M7W.json","view_paper":"https://pith.science/paper/DE527A2R","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2302.02971&json=true","fetch_graph":"https://pith.science/api/pith-number/DE527A2RJAHUUPRXJVQO4B5M7W/graph.json","fetch_events":"https://pith.science/api/pith-number/DE527A2RJAHUUPRXJVQO4B5M7W/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DE527A2RJAHUUPRXJVQO4B5M7W/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DE527A2RJAHUUPRXJVQO4B5M7W/action/storage_attestation","attest_author":"https://pith.science/pith/DE527A2RJAHUUPRXJVQO4B5M7W/action/author_attestation","sign_citation":"https://pith.science/pith/DE527A2RJAHUUPRXJVQO4B5M7W/action/citation_signature","submit_replication":"https://pith.science/pith/DE527A2RJAHUUPRXJVQO4B5M7W/action/replication_record"}},"created_at":"2026-07-05T05:39:08.731126+00:00","updated_at":"2026-07-05T05:39:08.731126+00:00"}