{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:BZGJMBWNVUF4AFHWAGRQ22MREC","short_pith_number":"pith:BZGJMBWN","schema_version":"1.0","canonical_sha256":"0e4c9606cdad0bc014f601a30d69912089da77c72ab91c4518f651c545edcc02","source":{"kind":"arxiv","id":"2402.03220","version":3},"attestation_state":"computed","paper":{"title":"The Benefits of Reusing Batches for Gradient Descent in Two-Layer Networks: Breaking the Curse of Information and Leap Exponents","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Emanuele Troiani, Florent Krzakala, Lenka Zdeborov\\'a, Luca Arnaboldi, Luca Pesce, Yatin Dandi","submitted_at":"2024-02-05T17:30:42Z","abstract_excerpt":"We investigate the training dynamics of two-layer neural networks when learning multi-index target functions. We focus on multi-pass gradient descent (GD) that reuses the batches multiple times and show that it significantly changes the conclusion about which functions are learnable compared to single-pass gradient descent. In particular, multi-pass GD with finite stepsize is found to overcome the limitations of gradient flow and single-pass GD given by the information exponent (Ben Arous et al., 2021) and leap exponent (Abbe et al., 2023) of the target function. We show that upon re-using bat"},"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":"2402.03220","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2024-02-05T17:30:42Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"ba2a1b2b6e52c0696fcca72aadccf1f313b34e7189ca452091a77ba532205e2f","abstract_canon_sha256":"bffc2d1885faa30b65d41c5e8b3a16e4e2b50241916ecd34db3a76af4bfb6cd2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:03:18.306151Z","signature_b64":"81JyjQIUisi7Xa+bqY6PCu0o+ItpNn3LABy2sx23fr53EvHwEHdMgyTihxY/aAJZwBUAu7kBbuMx7/Eq9Zq1CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0e4c9606cdad0bc014f601a30d69912089da77c72ab91c4518f651c545edcc02","last_reissued_at":"2026-07-05T09:03:18.305580Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:03:18.305580Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Benefits of Reusing Batches for Gradient Descent in Two-Layer Networks: Breaking the Curse of Information and Leap Exponents","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Emanuele Troiani, Florent Krzakala, Lenka Zdeborov\\'a, Luca Arnaboldi, Luca Pesce, Yatin Dandi","submitted_at":"2024-02-05T17:30:42Z","abstract_excerpt":"We investigate the training dynamics of two-layer neural networks when learning multi-index target functions. We focus on multi-pass gradient descent (GD) that reuses the batches multiple times and show that it significantly changes the conclusion about which functions are learnable compared to single-pass gradient descent. In particular, multi-pass GD with finite stepsize is found to overcome the limitations of gradient flow and single-pass GD given by the information exponent (Ben Arous et al., 2021) and leap exponent (Abbe et al., 2023) of the target function. We show that upon re-using bat"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.03220","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/2402.03220/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":"2402.03220","created_at":"2026-07-05T09:03:18.305638+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.03220v3","created_at":"2026-07-05T09:03:18.305638+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.03220","created_at":"2026-07-05T09:03:18.305638+00:00"},{"alias_kind":"pith_short_12","alias_value":"BZGJMBWNVUF4","created_at":"2026-07-05T09:03:18.305638+00:00"},{"alias_kind":"pith_short_16","alias_value":"BZGJMBWNVUF4AFHW","created_at":"2026-07-05T09:03:18.305638+00:00"},{"alias_kind":"pith_short_8","alias_value":"BZGJMBWN","created_at":"2026-07-05T09:03:18.305638+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.00207","citing_title":"Homogenization of $\\ell_2$-Adversarial Training in High-Dimensions: Exact Dynamics under Stochastic Gradient Descent","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2605.14567","citing_title":"Scaling Laws from Sequential Feature Recovery: A Solvable Hierarchical Model","ref_index":168,"is_internal_anchor":false},{"citing_arxiv_id":"2605.10237","citing_title":"The Benefits of Temporal Correlations: SGD Learns k-Juntas from Random Walks Efficiently","ref_index":126,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21691","citing_title":"There Will Be a Scientific Theory of Deep Learning","ref_index":58,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BZGJMBWNVUF4AFHWAGRQ22MREC","json":"https://pith.science/pith/BZGJMBWNVUF4AFHWAGRQ22MREC.json","graph_json":"https://pith.science/api/pith-number/BZGJMBWNVUF4AFHWAGRQ22MREC/graph.json","events_json":"https://pith.science/api/pith-number/BZGJMBWNVUF4AFHWAGRQ22MREC/events.json","paper":"https://pith.science/paper/BZGJMBWN"},"agent_actions":{"view_html":"https://pith.science/pith/BZGJMBWNVUF4AFHWAGRQ22MREC","download_json":"https://pith.science/pith/BZGJMBWNVUF4AFHWAGRQ22MREC.json","view_paper":"https://pith.science/paper/BZGJMBWN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.03220&json=true","fetch_graph":"https://pith.science/api/pith-number/BZGJMBWNVUF4AFHWAGRQ22MREC/graph.json","fetch_events":"https://pith.science/api/pith-number/BZGJMBWNVUF4AFHWAGRQ22MREC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BZGJMBWNVUF4AFHWAGRQ22MREC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BZGJMBWNVUF4AFHWAGRQ22MREC/action/storage_attestation","attest_author":"https://pith.science/pith/BZGJMBWNVUF4AFHWAGRQ22MREC/action/author_attestation","sign_citation":"https://pith.science/pith/BZGJMBWNVUF4AFHWAGRQ22MREC/action/citation_signature","submit_replication":"https://pith.science/pith/BZGJMBWNVUF4AFHWAGRQ22MREC/action/replication_record"}},"created_at":"2026-07-05T09:03:18.305638+00:00","updated_at":"2026-07-05T09:03:18.305638+00:00"}