{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:ZGCQEGVNFNBUBT4ULFUMAETJTW","short_pith_number":"pith:ZGCQEGVN","schema_version":"1.0","canonical_sha256":"c985021aad2b4340cf945968c012699d8b6d1ce958b14841a0e41ea507859f9e","source":{"kind":"arxiv","id":"1905.12558","version":3},"attestation_state":"computed","paper":{"title":"Limitations of the Empirical Fisher Approximation for Natural Gradient Descent","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Frederik Kunstner, Lukas Balles, Philipp Hennig","submitted_at":"2019-05-29T16:11:00Z","abstract_excerpt":"Natural gradient descent, which preconditions a gradient descent update with the Fisher information matrix of the underlying statistical model, is a way to capture partial second-order information. Several highly visible works have advocated an approximation known as the empirical Fisher, drawing connections between approximate second-order methods and heuristics like Adam. We dispute this argument by showing that the empirical Fisher---unlike the Fisher---does not generally capture second-order information. We further argue that the conditions under which the empirical Fisher approaches the F"},"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":"1905.12558","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2019-05-29T16:11:00Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"ab6cce0d964710cc12bc9f52e7a3480bb58b3a6aa54f33a7becb1e6c8ce48618","abstract_canon_sha256":"d67283b5fdba7c3e99803e3abf9923ab74a61f5626b811f229112b4f7b843092"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T01:08:46.502828Z","signature_b64":"SHfCds/EVS7I8cFV+Qd27zXj9PC3dc/1zs5/tMFhouOo4GVj60+4Xo1/2VC3jqPUV+tsiwp1DfM8Omv+uBnZBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c985021aad2b4340cf945968c012699d8b6d1ce958b14841a0e41ea507859f9e","last_reissued_at":"2026-07-05T01:08:46.502396Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T01:08:46.502396Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Limitations of the Empirical Fisher Approximation for Natural Gradient Descent","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Frederik Kunstner, Lukas Balles, Philipp Hennig","submitted_at":"2019-05-29T16:11:00Z","abstract_excerpt":"Natural gradient descent, which preconditions a gradient descent update with the Fisher information matrix of the underlying statistical model, is a way to capture partial second-order information. Several highly visible works have advocated an approximation known as the empirical Fisher, drawing connections between approximate second-order methods and heuristics like Adam. We dispute this argument by showing that the empirical Fisher---unlike the Fisher---does not generally capture second-order information. We further argue that the conditions under which the empirical Fisher approaches the F"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1905.12558","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/1905.12558/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":"1905.12558","created_at":"2026-07-05T01:08:46.502452+00:00"},{"alias_kind":"arxiv_version","alias_value":"1905.12558v3","created_at":"2026-07-05T01:08:46.502452+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1905.12558","created_at":"2026-07-05T01:08:46.502452+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZGCQEGVNFNBU","created_at":"2026-07-05T01:08:46.502452+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZGCQEGVNFNBUBT4U","created_at":"2026-07-05T01:08:46.502452+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZGCQEGVN","created_at":"2026-07-05T01:08:46.502452+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.21158","citing_title":"Dead-Direction Signatures: A Cheap Spectral Reading of Singular Complexity","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2606.19491","citing_title":"Algebraic Dead Directions in LayerNorm Transformers: A Forward-Pass-Only Diagnostic at LLM Scale","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05957","citing_title":"Dead Directions: Geometric Singular Learning","ref_index":24,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZGCQEGVNFNBUBT4ULFUMAETJTW","json":"https://pith.science/pith/ZGCQEGVNFNBUBT4ULFUMAETJTW.json","graph_json":"https://pith.science/api/pith-number/ZGCQEGVNFNBUBT4ULFUMAETJTW/graph.json","events_json":"https://pith.science/api/pith-number/ZGCQEGVNFNBUBT4ULFUMAETJTW/events.json","paper":"https://pith.science/paper/ZGCQEGVN"},"agent_actions":{"view_html":"https://pith.science/pith/ZGCQEGVNFNBUBT4ULFUMAETJTW","download_json":"https://pith.science/pith/ZGCQEGVNFNBUBT4ULFUMAETJTW.json","view_paper":"https://pith.science/paper/ZGCQEGVN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1905.12558&json=true","fetch_graph":"https://pith.science/api/pith-number/ZGCQEGVNFNBUBT4ULFUMAETJTW/graph.json","fetch_events":"https://pith.science/api/pith-number/ZGCQEGVNFNBUBT4ULFUMAETJTW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZGCQEGVNFNBUBT4ULFUMAETJTW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZGCQEGVNFNBUBT4ULFUMAETJTW/action/storage_attestation","attest_author":"https://pith.science/pith/ZGCQEGVNFNBUBT4ULFUMAETJTW/action/author_attestation","sign_citation":"https://pith.science/pith/ZGCQEGVNFNBUBT4ULFUMAETJTW/action/citation_signature","submit_replication":"https://pith.science/pith/ZGCQEGVNFNBUBT4ULFUMAETJTW/action/replication_record"}},"created_at":"2026-07-05T01:08:46.502452+00:00","updated_at":"2026-07-05T01:08:46.502452+00:00"}