{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:XORUO7RHA7YZO6BECWZIZS6PEI","short_pith_number":"pith:XORUO7RH","schema_version":"1.0","canonical_sha256":"bba3477e2707f197782415b28ccbcf222421da5b0cca1db4f718face4aeb3ca7","source":{"kind":"arxiv","id":"2404.04931","version":2},"attestation_state":"computed","paper":{"title":"The Sample Complexity of Gradient Descent in Stochastic Convex Optimization","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["math.OC"],"primary_cat":"cs.LG","authors_text":"Roi Livni","submitted_at":"2024-04-07T12:07:33Z","abstract_excerpt":"We analyze the sample complexity of full-batch Gradient Descent (GD) in the setup of non-smooth Stochastic Convex Optimization. We show that the generalization error of GD, with common choice of hyper-parameters, can be $\\tilde \\Theta(d/m + 1/\\sqrt{m})$, where $d$ is the dimension and $m$ is the sample size. This matches the sample complexity of \\emph{worst-case} empirical risk minimizers. That means that, in contrast with other algorithms, GD has no advantage over naive ERMs. Our bound follows from a new generalization bound that depends on both the dimension as well as the learning rate and "},"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":"2404.04931","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-04-07T12:07:33Z","cross_cats_sorted":["math.OC"],"title_canon_sha256":"881067d502118d9ae5fe3b3e832a55ad817bd7c15ec4bd84d5a157395c399289","abstract_canon_sha256":"bf47c04afcbc8b4d2cd8a4afd20acdd8d86bf272725f2e696a064307f528680e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:06:48.312588Z","signature_b64":"PsjrFSJq0q4JTUT9of9t4ghyPPN6xh6qiKxKXUzz9aDm121Jxb9m0WfSFr7Cvd7M1IU/FmD1Ta4v5i6b1A6eDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"bba3477e2707f197782415b28ccbcf222421da5b0cca1db4f718face4aeb3ca7","last_reissued_at":"2026-07-05T08:06:48.311992Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:06:48.311992Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Sample Complexity of Gradient Descent in Stochastic Convex Optimization","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["math.OC"],"primary_cat":"cs.LG","authors_text":"Roi Livni","submitted_at":"2024-04-07T12:07:33Z","abstract_excerpt":"We analyze the sample complexity of full-batch Gradient Descent (GD) in the setup of non-smooth Stochastic Convex Optimization. We show that the generalization error of GD, with common choice of hyper-parameters, can be $\\tilde \\Theta(d/m + 1/\\sqrt{m})$, where $d$ is the dimension and $m$ is the sample size. This matches the sample complexity of \\emph{worst-case} empirical risk minimizers. That means that, in contrast with other algorithms, GD has no advantage over naive ERMs. Our bound follows from a new generalization bound that depends on both the dimension as well as the learning rate and "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2404.04931","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/2404.04931/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":"2404.04931","created_at":"2026-07-05T08:06:48.312071+00:00"},{"alias_kind":"arxiv_version","alias_value":"2404.04931v2","created_at":"2026-07-05T08:06:48.312071+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2404.04931","created_at":"2026-07-05T08:06:48.312071+00:00"},{"alias_kind":"pith_short_12","alias_value":"XORUO7RHA7YZ","created_at":"2026-07-05T08:06:48.312071+00:00"},{"alias_kind":"pith_short_16","alias_value":"XORUO7RHA7YZO6BE","created_at":"2026-07-05T08:06:48.312071+00:00"},{"alias_kind":"pith_short_8","alias_value":"XORUO7RH","created_at":"2026-07-05T08:06:48.312071+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11431","citing_title":"Mirror Descent Beyond Euclidean Stability: An Exponential Separation in Initialization Sensitivity","ref_index":118,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/XORUO7RHA7YZO6BECWZIZS6PEI","json":"https://pith.science/pith/XORUO7RHA7YZO6BECWZIZS6PEI.json","graph_json":"https://pith.science/api/pith-number/XORUO7RHA7YZO6BECWZIZS6PEI/graph.json","events_json":"https://pith.science/api/pith-number/XORUO7RHA7YZO6BECWZIZS6PEI/events.json","paper":"https://pith.science/paper/XORUO7RH"},"agent_actions":{"view_html":"https://pith.science/pith/XORUO7RHA7YZO6BECWZIZS6PEI","download_json":"https://pith.science/pith/XORUO7RHA7YZO6BECWZIZS6PEI.json","view_paper":"https://pith.science/paper/XORUO7RH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2404.04931&json=true","fetch_graph":"https://pith.science/api/pith-number/XORUO7RHA7YZO6BECWZIZS6PEI/graph.json","fetch_events":"https://pith.science/api/pith-number/XORUO7RHA7YZO6BECWZIZS6PEI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/XORUO7RHA7YZO6BECWZIZS6PEI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/XORUO7RHA7YZO6BECWZIZS6PEI/action/storage_attestation","attest_author":"https://pith.science/pith/XORUO7RHA7YZO6BECWZIZS6PEI/action/author_attestation","sign_citation":"https://pith.science/pith/XORUO7RHA7YZO6BECWZIZS6PEI/action/citation_signature","submit_replication":"https://pith.science/pith/XORUO7RHA7YZO6BECWZIZS6PEI/action/replication_record"}},"created_at":"2026-07-05T08:06:48.312071+00:00","updated_at":"2026-07-05T08:06:48.312071+00:00"}