{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:4O4HQYDNFWQU6DQIXHXUHA3CP4","short_pith_number":"pith:4O4HQYDN","schema_version":"1.0","canonical_sha256":"e3b878606d2da14f0e08b9ef4383627f0d5e63b74c956d0f912f18c7fea0fd11","source":{"kind":"arxiv","id":"2107.03356","version":5},"attestation_state":"computed","paper":{"title":"M-FAC: Efficient Matrix-Free Approximations of Second-Order Information","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dan Alistarh, Eldar Kurtic, Elias Frantar","submitted_at":"2021-07-07T17:01:34Z","abstract_excerpt":"Efficiently approximating local curvature information of the loss function is a key tool for optimization and compression of deep neural networks. Yet, most existing methods to approximate second-order information have high computational or storage costs, which can limit their practicality. In this work, we investigate matrix-free, linear-time approaches for estimating Inverse-Hessian Vector Products (IHVPs) for the case when the Hessian can be approximated as a sum of rank-one matrices, as in the classic approximation of the Hessian by the empirical Fisher matrix. We propose two new algorithm"},"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":"2107.03356","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-07-07T17:01:34Z","cross_cats_sorted":[],"title_canon_sha256":"ae66a8cc4412c0d75dc1a996d4d1d97833a89069d0c734550ebc025cd83ba099","abstract_canon_sha256":"1bab612cd976770bd1da769a090dc2867d86b7b4c7f3740a6fba620f57eb23c4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:33:06.709706Z","signature_b64":"kv19edVVxf4S2bxKannsomSz+I1WxW9bQqug9JE5SLPHf8lHlXQdTHNdncg1pAXUKcr251P3+Kr1Fpx5hLW7DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e3b878606d2da14f0e08b9ef4383627f0d5e63b74c956d0f912f18c7fea0fd11","last_reissued_at":"2026-07-05T03:33:06.709070Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:33:06.709070Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"M-FAC: Efficient Matrix-Free Approximations of Second-Order Information","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Dan Alistarh, Eldar Kurtic, Elias Frantar","submitted_at":"2021-07-07T17:01:34Z","abstract_excerpt":"Efficiently approximating local curvature information of the loss function is a key tool for optimization and compression of deep neural networks. Yet, most existing methods to approximate second-order information have high computational or storage costs, which can limit their practicality. In this work, we investigate matrix-free, linear-time approaches for estimating Inverse-Hessian Vector Products (IHVPs) for the case when the Hessian can be approximated as a sum of rank-one matrices, as in the classic approximation of the Hessian by the empirical Fisher matrix. We propose two new algorithm"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2107.03356","kind":"arxiv","version":5},"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/2107.03356/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":"2107.03356","created_at":"2026-07-05T03:33:06.709140+00:00"},{"alias_kind":"arxiv_version","alias_value":"2107.03356v5","created_at":"2026-07-05T03:33:06.709140+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2107.03356","created_at":"2026-07-05T03:33:06.709140+00:00"},{"alias_kind":"pith_short_12","alias_value":"4O4HQYDNFWQU","created_at":"2026-07-05T03:33:06.709140+00:00"},{"alias_kind":"pith_short_16","alias_value":"4O4HQYDNFWQU6DQI","created_at":"2026-07-05T03:33:06.709140+00:00"},{"alias_kind":"pith_short_8","alias_value":"4O4HQYDN","created_at":"2026-07-05T03:33:06.709140+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2411.18376","citing_title":"Preserving Deep Representations In One-Shot Pruning: A Hessian-Free Second-Order Optimization Framework","ref_index":18,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4O4HQYDNFWQU6DQIXHXUHA3CP4","json":"https://pith.science/pith/4O4HQYDNFWQU6DQIXHXUHA3CP4.json","graph_json":"https://pith.science/api/pith-number/4O4HQYDNFWQU6DQIXHXUHA3CP4/graph.json","events_json":"https://pith.science/api/pith-number/4O4HQYDNFWQU6DQIXHXUHA3CP4/events.json","paper":"https://pith.science/paper/4O4HQYDN"},"agent_actions":{"view_html":"https://pith.science/pith/4O4HQYDNFWQU6DQIXHXUHA3CP4","download_json":"https://pith.science/pith/4O4HQYDNFWQU6DQIXHXUHA3CP4.json","view_paper":"https://pith.science/paper/4O4HQYDN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2107.03356&json=true","fetch_graph":"https://pith.science/api/pith-number/4O4HQYDNFWQU6DQIXHXUHA3CP4/graph.json","fetch_events":"https://pith.science/api/pith-number/4O4HQYDNFWQU6DQIXHXUHA3CP4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4O4HQYDNFWQU6DQIXHXUHA3CP4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4O4HQYDNFWQU6DQIXHXUHA3CP4/action/storage_attestation","attest_author":"https://pith.science/pith/4O4HQYDNFWQU6DQIXHXUHA3CP4/action/author_attestation","sign_citation":"https://pith.science/pith/4O4HQYDNFWQU6DQIXHXUHA3CP4/action/citation_signature","submit_replication":"https://pith.science/pith/4O4HQYDNFWQU6DQIXHXUHA3CP4/action/replication_record"}},"created_at":"2026-07-05T03:33:06.709140+00:00","updated_at":"2026-07-05T03:33:06.709140+00:00"}