{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:CSTLPE7G5Q2SCADUNMMW2DUAXS","short_pith_number":"pith:CSTLPE7G","schema_version":"1.0","canonical_sha256":"14a6b793e6ec352100746b196d0e80bc9894752ac8c06b0561f02c9e1913251d","source":{"kind":"arxiv","id":"2501.19099","version":2},"attestation_state":"computed","paper":{"title":"Elucidating Subspace Perturbation in Zeroth-Order Optimization: Theory and Practice at Scale","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Eunho Yang, Jihun Yun, Sihwan Park, Souvik Kundu, Sungyub Kim","submitted_at":"2025-01-31T12:46:04Z","abstract_excerpt":"Zeroth-order (ZO) optimization has emerged as a promising alternative to gradient-based backpropagation methods, particularly for black-box optimization and large language model (LLM) fine-tuning. However, ZO methods often suffer from slow convergence due to high-variance stochastic gradient estimators. While subspace perturbations, such as sparsity and low-rank constraints, have been explored to mitigate this issue, their effectiveness remains poorly understood. In this work, we develop a \\emph{unified theoretical framework} that analyzes both the convergence and generalization properties of "},"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":"2501.19099","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-01-31T12:46:04Z","cross_cats_sorted":[],"title_canon_sha256":"8d9d9b83b8abef4becab0785e2d29e3795008a7f7928c1adbd9f833a8562157b","abstract_canon_sha256":"1a665dd9931ccc74a495698d2962aba472cf407461a09e424ff6d8d1dd8cded8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:08:15.838860Z","signature_b64":"3lu1pTUMcZ/ww1rGG0HC21PKivfr5388eJg9zeFoH6MJ+esCsjts7dFFOifb6elzqQKqrTqiAatfFuR1n18yBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"14a6b793e6ec352100746b196d0e80bc9894752ac8c06b0561f02c9e1913251d","last_reissued_at":"2026-07-05T11:08:15.838236Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:08:15.838236Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Elucidating Subspace Perturbation in Zeroth-Order Optimization: Theory and Practice at Scale","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Eunho Yang, Jihun Yun, Sihwan Park, Souvik Kundu, Sungyub Kim","submitted_at":"2025-01-31T12:46:04Z","abstract_excerpt":"Zeroth-order (ZO) optimization has emerged as a promising alternative to gradient-based backpropagation methods, particularly for black-box optimization and large language model (LLM) fine-tuning. However, ZO methods often suffer from slow convergence due to high-variance stochastic gradient estimators. While subspace perturbations, such as sparsity and low-rank constraints, have been explored to mitigate this issue, their effectiveness remains poorly understood. In this work, we develop a \\emph{unified theoretical framework} that analyzes both the convergence and generalization properties of "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.19099","kind":"arxiv","version":2},"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/2501.19099/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":"2501.19099","created_at":"2026-07-05T11:08:15.838325+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.19099v2","created_at":"2026-07-05T11:08:15.838325+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.19099","created_at":"2026-07-05T11:08:15.838325+00:00"},{"alias_kind":"pith_short_12","alias_value":"CSTLPE7G5Q2S","created_at":"2026-07-05T11:08:15.838325+00:00"},{"alias_kind":"pith_short_16","alias_value":"CSTLPE7G5Q2SCADU","created_at":"2026-07-05T11:08:15.838325+00:00"},{"alias_kind":"pith_short_8","alias_value":"CSTLPE7G","created_at":"2026-07-05T11:08:15.838325+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.15622","citing_title":"Position: Zeroth-Order Optimization in Deep Learning Is Underexplored, Not Underpowered","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2511.02757","citing_title":"ConMeZO: Adaptive Descent-Direction Sampling for Gradient-Free Finetuning of Large Language Models","ref_index":11,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/CSTLPE7G5Q2SCADUNMMW2DUAXS","json":"https://pith.science/pith/CSTLPE7G5Q2SCADUNMMW2DUAXS.json","graph_json":"https://pith.science/api/pith-number/CSTLPE7G5Q2SCADUNMMW2DUAXS/graph.json","events_json":"https://pith.science/api/pith-number/CSTLPE7G5Q2SCADUNMMW2DUAXS/events.json","paper":"https://pith.science/paper/CSTLPE7G"},"agent_actions":{"view_html":"https://pith.science/pith/CSTLPE7G5Q2SCADUNMMW2DUAXS","download_json":"https://pith.science/pith/CSTLPE7G5Q2SCADUNMMW2DUAXS.json","view_paper":"https://pith.science/paper/CSTLPE7G","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.19099&json=true","fetch_graph":"https://pith.science/api/pith-number/CSTLPE7G5Q2SCADUNMMW2DUAXS/graph.json","fetch_events":"https://pith.science/api/pith-number/CSTLPE7G5Q2SCADUNMMW2DUAXS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/CSTLPE7G5Q2SCADUNMMW2DUAXS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/CSTLPE7G5Q2SCADUNMMW2DUAXS/action/storage_attestation","attest_author":"https://pith.science/pith/CSTLPE7G5Q2SCADUNMMW2DUAXS/action/author_attestation","sign_citation":"https://pith.science/pith/CSTLPE7G5Q2SCADUNMMW2DUAXS/action/citation_signature","submit_replication":"https://pith.science/pith/CSTLPE7G5Q2SCADUNMMW2DUAXS/action/replication_record"}},"created_at":"2026-07-05T11:08:15.838325+00:00","updated_at":"2026-07-05T11:08:15.838325+00:00"}