{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:LRMUNZCNPIMY4V2T2FGVQ2ERT6","short_pith_number":"pith:LRMUNZCN","schema_version":"1.0","canonical_sha256":"5c5946e44d7a198e5753d14d5868919f97a3da63abd0d8af8264f8f964eb8bbd","source":{"kind":"arxiv","id":"2205.11787","version":3},"attestation_state":"computed","paper":{"title":"Quadratic models for understanding catapult dynamics of neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Adityanarayanan Radhakrishnan, Chaoyue Liu, Libin Zhu, Mikhail Belkin","submitted_at":"2022-05-24T05:03:06Z","abstract_excerpt":"While neural networks can be approximated by linear models as their width increases, certain properties of wide neural networks cannot be captured by linear models. In this work we show that recently proposed Neural Quadratic Models can exhibit the \"catapult phase\" [Lewkowycz et al. 2020] that arises when training such models with large learning rates. We then empirically show that the behaviour of neural quadratic models parallels that of neural networks in generalization, especially in the catapult phase regime. Our analysis further demonstrates that quadratic models can be an effective tool"},"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":"2205.11787","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2022-05-24T05:03:06Z","cross_cats_sorted":["math.OC","stat.ML"],"title_canon_sha256":"56e46c9a073dc31c3cff7c94f06cbba308bff8cb86421c268b58c110ea7059fe","abstract_canon_sha256":"fd36327f93ebed507c90226468b4ce2c61ad41bce516f794676f6f21d224a707"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:14:16.801560Z","signature_b64":"rIgfKGWMnjzv2dREZfTyZpeAt7CHhmjBjxCngs+D58XjjmIk5vPCBMWcx5WVVRGY2IKpfqaOt1pxLDdF8sv2DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5c5946e44d7a198e5753d14d5868919f97a3da63abd0d8af8264f8f964eb8bbd","last_reissued_at":"2026-07-05T08:14:16.801089Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:14:16.801089Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Quadratic models for understanding catapult dynamics of neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["math.OC","stat.ML"],"primary_cat":"cs.LG","authors_text":"Adityanarayanan Radhakrishnan, Chaoyue Liu, Libin Zhu, Mikhail Belkin","submitted_at":"2022-05-24T05:03:06Z","abstract_excerpt":"While neural networks can be approximated by linear models as their width increases, certain properties of wide neural networks cannot be captured by linear models. In this work we show that recently proposed Neural Quadratic Models can exhibit the \"catapult phase\" [Lewkowycz et al. 2020] that arises when training such models with large learning rates. We then empirically show that the behaviour of neural quadratic models parallels that of neural networks in generalization, especially in the catapult phase regime. Our analysis further demonstrates that quadratic models can be an effective tool"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.11787","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/2205.11787/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":"2205.11787","created_at":"2026-07-05T08:14:16.801141+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.11787v3","created_at":"2026-07-05T08:14:16.801141+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.11787","created_at":"2026-07-05T08:14:16.801141+00:00"},{"alias_kind":"pith_short_12","alias_value":"LRMUNZCNPIMY","created_at":"2026-07-05T08:14:16.801141+00:00"},{"alias_kind":"pith_short_16","alias_value":"LRMUNZCNPIMY4V2T","created_at":"2026-07-05T08:14:16.801141+00:00"},{"alias_kind":"pith_short_8","alias_value":"LRMUNZCN","created_at":"2026-07-05T08:14:16.801141+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2509.08233","citing_title":"Strategies for Improving Communication Efficiency in Distributed and Federated Learning: Compression, Local Training, and Personalization","ref_index":254,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/LRMUNZCNPIMY4V2T2FGVQ2ERT6","json":"https://pith.science/pith/LRMUNZCNPIMY4V2T2FGVQ2ERT6.json","graph_json":"https://pith.science/api/pith-number/LRMUNZCNPIMY4V2T2FGVQ2ERT6/graph.json","events_json":"https://pith.science/api/pith-number/LRMUNZCNPIMY4V2T2FGVQ2ERT6/events.json","paper":"https://pith.science/paper/LRMUNZCN"},"agent_actions":{"view_html":"https://pith.science/pith/LRMUNZCNPIMY4V2T2FGVQ2ERT6","download_json":"https://pith.science/pith/LRMUNZCNPIMY4V2T2FGVQ2ERT6.json","view_paper":"https://pith.science/paper/LRMUNZCN","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.11787&json=true","fetch_graph":"https://pith.science/api/pith-number/LRMUNZCNPIMY4V2T2FGVQ2ERT6/graph.json","fetch_events":"https://pith.science/api/pith-number/LRMUNZCNPIMY4V2T2FGVQ2ERT6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LRMUNZCNPIMY4V2T2FGVQ2ERT6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LRMUNZCNPIMY4V2T2FGVQ2ERT6/action/storage_attestation","attest_author":"https://pith.science/pith/LRMUNZCNPIMY4V2T2FGVQ2ERT6/action/author_attestation","sign_citation":"https://pith.science/pith/LRMUNZCNPIMY4V2T2FGVQ2ERT6/action/citation_signature","submit_replication":"https://pith.science/pith/LRMUNZCNPIMY4V2T2FGVQ2ERT6/action/replication_record"}},"created_at":"2026-07-05T08:14:16.801141+00:00","updated_at":"2026-07-05T08:14:16.801141+00:00"}