{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:PW5LW4TR7CZCLEAEL5M5EEHMCK","short_pith_number":"pith:PW5LW4TR","schema_version":"1.0","canonical_sha256":"7dbabb7271f8b22590045f59d210ec12ac42949c120d3f721c5aa58f0ad7d010","source":{"kind":"arxiv","id":"2605.13913","version":1},"attestation_state":"computed","paper":{"title":"A Survey on Data-Dependent Worst-Case Generalization Bounds","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A single template inequality unifies data-dependent generalization bounds for overparameterized networks.","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Hubert Leroux, Jean Marcus, Julien Roger","submitted_at":"2026-05-13T09:03:50Z","abstract_excerpt":"Deep neural networks generalize well despite being heavily overparameterized, in apparent contradiction with classical learning theory based on uniform convergence over fixed hypothesis spaces. Uniform bounds over the entire parameter space are vacuous in this regime, and recent work has shown that non-vacuous guarantees can be recovered by restricting attention to the part of parameter space that the algorithm actually visits. This survey paper organizes this line of work around three steps: extending PAC-Bayesian theory to random, data-dependent hypothesis sets (arXiv:2404.17442); refining t"},"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":true,"formal_links_present":false},"canonical_record":{"source":{"id":"2605.13913","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2026-05-13T09:03:50Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"1a57654a3d6ac3b85842c537c0c88c9f5be87d0cd3b654710746d2dc72cd795e","abstract_canon_sha256":"cc1291814cfd0e0fe894ed683b36851823512d366a8263ff2c4e8c35b857372f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:18.789466Z","signature_b64":"qSK/lUTtUUrJpDcQQr1VWyM4Ebul0Ez6c6XariAXhK2+42Ss8hRwlJfDKmq0AggP4SPc5AD7fFPzllwoE3oyAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7dbabb7271f8b22590045f59d210ec12ac42949c120d3f721c5aa58f0ad7d010","last_reissued_at":"2026-05-17T23:39:18.788731Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:18.788731Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Survey on Data-Dependent Worst-Case Generalization Bounds","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A single template inequality unifies data-dependent generalization bounds for overparameterized networks.","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Hubert Leroux, Jean Marcus, Julien Roger","submitted_at":"2026-05-13T09:03:50Z","abstract_excerpt":"Deep neural networks generalize well despite being heavily overparameterized, in apparent contradiction with classical learning theory based on uniform convergence over fixed hypothesis spaces. Uniform bounds over the entire parameter space are vacuous in this regime, and recent work has shown that non-vacuous guarantees can be recovered by restricting attention to the part of parameter space that the algorithm actually visits. This survey paper organizes this line of work around three steps: extending PAC-Bayesian theory to random, data-dependent hypothesis sets (arXiv:2404.17442); refining t"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We unify these contributions around a single template inequality and a head-to-head comparison of the resulting bounds.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The survey assumes that the cited works' individual bounds remain valid when placed inside the common template; any hidden incompatibility between the data-dependent set constructions, geometric descriptors, and stability assumptions would invalidate the unification.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"The survey unifies extensions of PAC-Bayesian theory to data-dependent sets, geometric and topological complexity measures of optimization trajectories, and stability replacements for information terms into one template inequality with comparative evaluation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A single template inequality unifies data-dependent generalization bounds for overparameterized networks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"15f9f87f9fe57a7885913b0cd38c397e05473e9d489d423640ae6b94c87d1ea0"},"source":{"id":"2605.13913","kind":"arxiv","version":1},"verdict":{"id":"c1ff588f-2765-453a-b521-a135b7cf29ea","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:57:57.302874Z","strongest_claim":"We unify these contributions around a single template inequality and a head-to-head comparison of the resulting bounds.","one_line_summary":"The survey unifies extensions of PAC-Bayesian theory to data-dependent sets, geometric and topological complexity measures of optimization trajectories, and stability replacements for information terms into one template inequality with comparative evaluation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The survey assumes that the cited works' individual bounds remain valid when placed inside the common template; any hidden incompatibility between the data-dependent set constructions, geometric descriptors, and stability assumptions would invalidate the unification.","pith_extraction_headline":"A single template inequality unifies data-dependent generalization bounds for overparameterized networks."},"references":{"count":17,"sample":[{"doi":"","year":2009,"title":"Neural network learning: Theoretical foundations , author=. 2009 , publisher=","work_id":"d46f0bc6-2131-4fbf-8671-3eaa5347ae17","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Advances in Neural Information Processing Systems , volume=","work_id":"3fd6b321-7b63-4afa-8b28-5b9040b2703c","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Journal of machine learning research , volume=","work_id":"82cea0e0-0fe5-47e4-8caa-eb6c64eaaee5","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"A PAC-Bayesian approach to adaptive classification , author=. preprint , volume=","work_id":"eb161c7f-a4f5-4fb6-ad71-229579e73d0d","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Proceedings of the 26th Annual International Conference on Machine Learning , pages=","work_id":"6cebc8d8-33d5-4eac-b210-5e20432c4704","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":17,"snapshot_sha256":"cce1bbbd05dad033713d2e3ad3377611d8e79d06b110a0d87c61863abde6f185","internal_anchors":2},"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":"2605.13913","created_at":"2026-05-17T23:39:18.788862+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.13913v1","created_at":"2026-05-17T23:39:18.788862+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13913","created_at":"2026-05-17T23:39:18.788862+00:00"},{"alias_kind":"pith_short_12","alias_value":"PW5LW4TR7CZC","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_16","alias_value":"PW5LW4TR7CZCLEAE","created_at":"2026-05-18T12:33:37.589309+00:00"},{"alias_kind":"pith_short_8","alias_value":"PW5LW4TR","created_at":"2026-05-18T12:33:37.589309+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PW5LW4TR7CZCLEAEL5M5EEHMCK","json":"https://pith.science/pith/PW5LW4TR7CZCLEAEL5M5EEHMCK.json","graph_json":"https://pith.science/api/pith-number/PW5LW4TR7CZCLEAEL5M5EEHMCK/graph.json","events_json":"https://pith.science/api/pith-number/PW5LW4TR7CZCLEAEL5M5EEHMCK/events.json","paper":"https://pith.science/paper/PW5LW4TR"},"agent_actions":{"view_html":"https://pith.science/pith/PW5LW4TR7CZCLEAEL5M5EEHMCK","download_json":"https://pith.science/pith/PW5LW4TR7CZCLEAEL5M5EEHMCK.json","view_paper":"https://pith.science/paper/PW5LW4TR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.13913&json=true","fetch_graph":"https://pith.science/api/pith-number/PW5LW4TR7CZCLEAEL5M5EEHMCK/graph.json","fetch_events":"https://pith.science/api/pith-number/PW5LW4TR7CZCLEAEL5M5EEHMCK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PW5LW4TR7CZCLEAEL5M5EEHMCK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PW5LW4TR7CZCLEAEL5M5EEHMCK/action/storage_attestation","attest_author":"https://pith.science/pith/PW5LW4TR7CZCLEAEL5M5EEHMCK/action/author_attestation","sign_citation":"https://pith.science/pith/PW5LW4TR7CZCLEAEL5M5EEHMCK/action/citation_signature","submit_replication":"https://pith.science/pith/PW5LW4TR7CZCLEAEL5M5EEHMCK/action/replication_record"}},"created_at":"2026-05-17T23:39:18.788862+00:00","updated_at":"2026-05-17T23:39:18.788862+00:00"}