{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:3JNLGQBZNTTPZZARRSBLEXXGGQ","short_pith_number":"pith:3JNLGQBZ","schema_version":"1.0","canonical_sha256":"da5ab340396ce6fce4118c82b25ee6342eb3eae9837668258ced95ef19687268","source":{"kind":"arxiv","id":"2305.18502","version":2},"attestation_state":"computed","paper":{"title":"Escaping mediocrity: how two-layer networks learn hard generalized linear models with SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Bruno Loureiro, Florent Krzakala, Luca Arnaboldi, Ludovic Stephan","submitted_at":"2023-05-29T14:40:56Z","abstract_excerpt":"This study explores the sample complexity for two-layer neural networks to learn a generalized linear target function under Stochastic Gradient Descent (SGD), focusing on the challenging regime where many flat directions are present at initialization. It is well-established that in this scenario $n=O(d \\log d)$ samples are typically needed. However, we provide precise results concerning the pre-factors in high-dimensional contexts and for varying widths. Notably, our findings suggest that overparameterization can only enhance convergence by a constant factor within this problem class. These in"},"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":"2305.18502","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2023-05-29T14:40:56Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"c3422caa45f25052b14d26b31d1f854e13d3ab785aad226add5befaf4caa1dfa","abstract_canon_sha256":"c220bc74e99062bb8896016fb0591632addc43a3998af6614741cbf3131b370b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:50:50.085229Z","signature_b64":"gihol9NlVEVMT9mhN8U+JuvxTYNSziHjbHuwT1iE/EcA+t4GEq1Kc0MKbV718MAFEJYTgdhFPWZC25YeYu/0AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"da5ab340396ce6fce4118c82b25ee6342eb3eae9837668258ced95ef19687268","last_reissued_at":"2026-07-05T07:50:50.084721Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:50:50.084721Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Escaping mediocrity: how two-layer networks learn hard generalized linear models with SGD","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Bruno Loureiro, Florent Krzakala, Luca Arnaboldi, Ludovic Stephan","submitted_at":"2023-05-29T14:40:56Z","abstract_excerpt":"This study explores the sample complexity for two-layer neural networks to learn a generalized linear target function under Stochastic Gradient Descent (SGD), focusing on the challenging regime where many flat directions are present at initialization. It is well-established that in this scenario $n=O(d \\log d)$ samples are typically needed. However, we provide precise results concerning the pre-factors in high-dimensional contexts and for varying widths. Notably, our findings suggest that overparameterization can only enhance convergence by a constant factor within this problem class. These in"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.18502","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/2305.18502/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":"2305.18502","created_at":"2026-07-05T07:50:50.084782+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.18502v2","created_at":"2026-07-05T07:50:50.084782+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.18502","created_at":"2026-07-05T07:50:50.084782+00:00"},{"alias_kind":"pith_short_12","alias_value":"3JNLGQBZNTTP","created_at":"2026-07-05T07:50:50.084782+00:00"},{"alias_kind":"pith_short_16","alias_value":"3JNLGQBZNTTPZZAR","created_at":"2026-07-05T07:50:50.084782+00:00"},{"alias_kind":"pith_short_8","alias_value":"3JNLGQBZ","created_at":"2026-07-05T07:50:50.084782+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"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":2,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01288","citing_title":"A Theory of Saddle Escape in Deep Nonlinear Networks","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2604.03068","citing_title":"Escape dynamics and implicit bias of one-pass SGD in overparameterized quadratic networks","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01288","citing_title":"A Theory of Saddle Escape in Deep Nonlinear Networks","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2605.01288","citing_title":"A Theory of Saddle Escape in Deep Nonlinear Networks","ref_index":4,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3JNLGQBZNTTPZZARRSBLEXXGGQ","json":"https://pith.science/pith/3JNLGQBZNTTPZZARRSBLEXXGGQ.json","graph_json":"https://pith.science/api/pith-number/3JNLGQBZNTTPZZARRSBLEXXGGQ/graph.json","events_json":"https://pith.science/api/pith-number/3JNLGQBZNTTPZZARRSBLEXXGGQ/events.json","paper":"https://pith.science/paper/3JNLGQBZ"},"agent_actions":{"view_html":"https://pith.science/pith/3JNLGQBZNTTPZZARRSBLEXXGGQ","download_json":"https://pith.science/pith/3JNLGQBZNTTPZZARRSBLEXXGGQ.json","view_paper":"https://pith.science/paper/3JNLGQBZ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.18502&json=true","fetch_graph":"https://pith.science/api/pith-number/3JNLGQBZNTTPZZARRSBLEXXGGQ/graph.json","fetch_events":"https://pith.science/api/pith-number/3JNLGQBZNTTPZZARRSBLEXXGGQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3JNLGQBZNTTPZZARRSBLEXXGGQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3JNLGQBZNTTPZZARRSBLEXXGGQ/action/storage_attestation","attest_author":"https://pith.science/pith/3JNLGQBZNTTPZZARRSBLEXXGGQ/action/author_attestation","sign_citation":"https://pith.science/pith/3JNLGQBZNTTPZZARRSBLEXXGGQ/action/citation_signature","submit_replication":"https://pith.science/pith/3JNLGQBZNTTPZZARRSBLEXXGGQ/action/replication_record"}},"created_at":"2026-07-05T07:50:50.084782+00:00","updated_at":"2026-07-05T07:50:50.084782+00:00"}