{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:VPAX5PPWXEVIG4KSNW347MF5AA","short_pith_number":"pith:VPAX5PPW","schema_version":"1.0","canonical_sha256":"abc17ebdf6b92a8371526db7cfb0bd0027b7469b3c2ceea0d46640566cd38a92","source":{"kind":"arxiv","id":"2405.13817","version":1},"attestation_state":"computed","paper":{"title":"Thermodynamic Natural Gradient Descent","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.ET"],"primary_cat":"cs.LG","authors_text":"Denis Melanson, Gavin Crooks, Kaelan Donatella, Maxwell Aifer, Patrick J. Coles, Samuel Duffield","submitted_at":"2024-05-22T16:47:03Z","abstract_excerpt":"Second-order training methods have better convergence properties than gradient descent but are rarely used in practice for large-scale training due to their computational overhead. This can be viewed as a hardware limitation (imposed by digital computers). Here we show that natural gradient descent (NGD), a second-order method, can have a similar computational complexity per iteration to a first-order method, when employing appropriate hardware. We present a new hybrid digital-analog algorithm for training neural networks that is equivalent to NGD in a certain parameter regime but avoids prohi"},"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":"2405.13817","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-05-22T16:47:03Z","cross_cats_sorted":["cs.ET"],"title_canon_sha256":"7421488d9cfaf5d481def2a508f1a5660831c594ef5623b566c6bd9b25ef16af","abstract_canon_sha256":"e5fbaff3dcacd0049266903b81d9e011913a3a4999290741060e0761777e7dba"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:22:07.339143Z","signature_b64":"SkcDgzmoAu+aV9RfqM8zbZj1Q68Z0RAWoKhQKFF0YvWbfDVewlnsX6hFAdskr4VuDWaF3ne+FGqeB5cT1BcfBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"abc17ebdf6b92a8371526db7cfb0bd0027b7469b3c2ceea0d46640566cd38a92","last_reissued_at":"2026-07-05T08:22:07.338616Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:22:07.338616Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Thermodynamic Natural Gradient Descent","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.ET"],"primary_cat":"cs.LG","authors_text":"Denis Melanson, Gavin Crooks, Kaelan Donatella, Maxwell Aifer, Patrick J. Coles, Samuel Duffield","submitted_at":"2024-05-22T16:47:03Z","abstract_excerpt":"Second-order training methods have better convergence properties than gradient descent but are rarely used in practice for large-scale training due to their computational overhead. This can be viewed as a hardware limitation (imposed by digital computers). Here we show that natural gradient descent (NGD), a second-order method, can have a similar computational complexity per iteration to a first-order method, when employing appropriate hardware. We present a new hybrid digital-analog algorithm for training neural networks that is equivalent to NGD in a certain parameter regime but avoids prohi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2405.13817","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/2405.13817/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":"2405.13817","created_at":"2026-07-05T08:22:07.338683+00:00"},{"alias_kind":"arxiv_version","alias_value":"2405.13817v1","created_at":"2026-07-05T08:22:07.338683+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2405.13817","created_at":"2026-07-05T08:22:07.338683+00:00"},{"alias_kind":"pith_short_12","alias_value":"VPAX5PPWXEVI","created_at":"2026-07-05T08:22:07.338683+00:00"},{"alias_kind":"pith_short_16","alias_value":"VPAX5PPWXEVIG4KS","created_at":"2026-07-05T08:22:07.338683+00:00"},{"alias_kind":"pith_short_8","alias_value":"VPAX5PPW","created_at":"2026-07-05T08:22:07.338683+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.13179","citing_title":"Modern analog computing for solving differential and matrix equations","ref_index":69,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VPAX5PPWXEVIG4KSNW347MF5AA","json":"https://pith.science/pith/VPAX5PPWXEVIG4KSNW347MF5AA.json","graph_json":"https://pith.science/api/pith-number/VPAX5PPWXEVIG4KSNW347MF5AA/graph.json","events_json":"https://pith.science/api/pith-number/VPAX5PPWXEVIG4KSNW347MF5AA/events.json","paper":"https://pith.science/paper/VPAX5PPW"},"agent_actions":{"view_html":"https://pith.science/pith/VPAX5PPWXEVIG4KSNW347MF5AA","download_json":"https://pith.science/pith/VPAX5PPWXEVIG4KSNW347MF5AA.json","view_paper":"https://pith.science/paper/VPAX5PPW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2405.13817&json=true","fetch_graph":"https://pith.science/api/pith-number/VPAX5PPWXEVIG4KSNW347MF5AA/graph.json","fetch_events":"https://pith.science/api/pith-number/VPAX5PPWXEVIG4KSNW347MF5AA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VPAX5PPWXEVIG4KSNW347MF5AA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VPAX5PPWXEVIG4KSNW347MF5AA/action/storage_attestation","attest_author":"https://pith.science/pith/VPAX5PPWXEVIG4KSNW347MF5AA/action/author_attestation","sign_citation":"https://pith.science/pith/VPAX5PPWXEVIG4KSNW347MF5AA/action/citation_signature","submit_replication":"https://pith.science/pith/VPAX5PPWXEVIG4KSNW347MF5AA/action/replication_record"}},"created_at":"2026-07-05T08:22:07.338683+00:00","updated_at":"2026-07-05T08:22:07.338683+00:00"}