{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:KUIV43KVG6RPI7XZTFEO4L6O2V","short_pith_number":"pith:KUIV43KV","schema_version":"1.0","canonical_sha256":"55115e6d5537a2f47ef99948ee2fced56941e4ce0a811d4d3f401e0d472eb071","source":{"kind":"arxiv","id":"2312.05705","version":4},"attestation_state":"computed","paper":{"title":"Structured Inverse-Free Natural Gradient: Memory-Efficient & Numerically-Stable KFAC","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Agustinus Kristiadi, Alireza Makhzani, Felix Dangel, Kirill Neklyudov, Richard E. Turner, Runa Eschenhagen, Wu Lin","submitted_at":"2023-12-09T23:13:32Z","abstract_excerpt":"Second-order methods such as KFAC can be useful for neural net training. However, they are often memory-inefficient since their preconditioning Kronecker factors are dense, and numerically unstable in low precision as they require matrix inversion or decomposition. These limitations render such methods unpopular for modern mixed-precision training. We address them by (i) formulating an inverse-free KFAC update and (ii) imposing structures in the Kronecker factors, resulting in structured inverse-free natural gradient descent (SINGD). On modern neural networks, we show that SINGD is memory-effi"},"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":"2312.05705","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2023-12-09T23:13:32Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"4dc785090d04ae403bbdf03f232762c451b93278873f8e1099d8fe17a619bcc4","abstract_canon_sha256":"1b8c1480e16f7bb88285ed960bc9a42e6cb612a6620a3501478ab8a4fc7888af"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:47:09.186045Z","signature_b64":"PM60MbgDLUwEykWYk0ixVXEfjX6V30gPw53arVDLQyFBwqpYdRUPOPgFbtWgJrxdDELGJ4p+ojsN77ogIHSmAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"55115e6d5537a2f47ef99948ee2fced56941e4ce0a811d4d3f401e0d472eb071","last_reissued_at":"2026-07-05T08:47:09.185305Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:47:09.185305Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Structured Inverse-Free Natural Gradient: Memory-Efficient & Numerically-Stable KFAC","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Agustinus Kristiadi, Alireza Makhzani, Felix Dangel, Kirill Neklyudov, Richard E. Turner, Runa Eschenhagen, Wu Lin","submitted_at":"2023-12-09T23:13:32Z","abstract_excerpt":"Second-order methods such as KFAC can be useful for neural net training. However, they are often memory-inefficient since their preconditioning Kronecker factors are dense, and numerically unstable in low precision as they require matrix inversion or decomposition. These limitations render such methods unpopular for modern mixed-precision training. We address them by (i) formulating an inverse-free KFAC update and (ii) imposing structures in the Kronecker factors, resulting in structured inverse-free natural gradient descent (SINGD). On modern neural networks, we show that SINGD is memory-effi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2312.05705","kind":"arxiv","version":4},"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/2312.05705/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":"2312.05705","created_at":"2026-07-05T08:47:09.185429+00:00"},{"alias_kind":"arxiv_version","alias_value":"2312.05705v4","created_at":"2026-07-05T08:47:09.185429+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2312.05705","created_at":"2026-07-05T08:47:09.185429+00:00"},{"alias_kind":"pith_short_12","alias_value":"KUIV43KVG6RP","created_at":"2026-07-05T08:47:09.185429+00:00"},{"alias_kind":"pith_short_16","alias_value":"KUIV43KVG6RPI7XZ","created_at":"2026-07-05T08:47:09.185429+00:00"},{"alias_kind":"pith_short_8","alias_value":"KUIV43KV","created_at":"2026-07-05T08:47:09.185429+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.06418","citing_title":"Double Preconditioning (DoPr): Optimization for Test-Time Performance, not Validation Loss","ref_index":83,"is_internal_anchor":false},{"citing_arxiv_id":"2606.30813","citing_title":"Gradient Smoothing: Coupling Layer-wise Updates for Improved Optimization","ref_index":19,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KUIV43KVG6RPI7XZTFEO4L6O2V","json":"https://pith.science/pith/KUIV43KVG6RPI7XZTFEO4L6O2V.json","graph_json":"https://pith.science/api/pith-number/KUIV43KVG6RPI7XZTFEO4L6O2V/graph.json","events_json":"https://pith.science/api/pith-number/KUIV43KVG6RPI7XZTFEO4L6O2V/events.json","paper":"https://pith.science/paper/KUIV43KV"},"agent_actions":{"view_html":"https://pith.science/pith/KUIV43KVG6RPI7XZTFEO4L6O2V","download_json":"https://pith.science/pith/KUIV43KVG6RPI7XZTFEO4L6O2V.json","view_paper":"https://pith.science/paper/KUIV43KV","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2312.05705&json=true","fetch_graph":"https://pith.science/api/pith-number/KUIV43KVG6RPI7XZTFEO4L6O2V/graph.json","fetch_events":"https://pith.science/api/pith-number/KUIV43KVG6RPI7XZTFEO4L6O2V/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KUIV43KVG6RPI7XZTFEO4L6O2V/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KUIV43KVG6RPI7XZTFEO4L6O2V/action/storage_attestation","attest_author":"https://pith.science/pith/KUIV43KVG6RPI7XZTFEO4L6O2V/action/author_attestation","sign_citation":"https://pith.science/pith/KUIV43KVG6RPI7XZTFEO4L6O2V/action/citation_signature","submit_replication":"https://pith.science/pith/KUIV43KVG6RPI7XZTFEO4L6O2V/action/replication_record"}},"created_at":"2026-07-05T08:47:09.185429+00:00","updated_at":"2026-07-05T08:47:09.185429+00:00"}