{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:JR7N5D5K3HXQYSCUBMNG3CZINO","short_pith_number":"pith:JR7N5D5K","schema_version":"1.0","canonical_sha256":"4c7ede8faad9ef0c48540b1a6d8b286ba55fad1783aa19cefa05154de111b3fe","source":{"kind":"arxiv","id":"2511.01938","version":3},"attestation_state":"computed","paper":{"title":"The Geometry of Grokking: Norm Minimization on the Zero-Loss Manifold","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Tiberiu Musat","submitted_at":"2025-11-02T18:44:42Z","abstract_excerpt":"Grokking is a puzzling phenomenon in neural networks where full generalization occurs only after a substantial delay following the complete memorization of the training data. Previous research has linked this delayed generalization to representation learning driven by weight decay, but the precise underlying dynamics remain elusive. In this paper, we argue that post-memorization learning can be understood through the lens of constrained optimization: gradient descent effectively minimizes the weight norm on the zero-loss manifold. We formally prove this in the limit of infinitesimally small le"},"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":"2511.01938","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-11-02T18:44:42Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2c5aa44a1072d3a580eb069994efa8f9d87de586c14d29289b081684fbb0021c","abstract_canon_sha256":"f52998b077ac93c46a17a74fffb6b62c861cc353363c4daa4e6810b30277cadf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-02T01:03:37.873712Z","signature_b64":"JVUMKE0C/DeVFrQM0RhctaMoxzNHUR6IwTWgapBHbZZw1Ko1ZLXvWy1n6q6SGtLxDd/Iwrw7BJWrh8zHIzcMBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4c7ede8faad9ef0c48540b1a6d8b286ba55fad1783aa19cefa05154de111b3fe","last_reissued_at":"2026-06-02T01:03:37.873161Z","signature_status":"signed_v1","first_computed_at":"2026-06-02T01:03:37.873161Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"The Geometry of Grokking: Norm Minimization on the Zero-Loss Manifold","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Tiberiu Musat","submitted_at":"2025-11-02T18:44:42Z","abstract_excerpt":"Grokking is a puzzling phenomenon in neural networks where full generalization occurs only after a substantial delay following the complete memorization of the training data. Previous research has linked this delayed generalization to representation learning driven by weight decay, but the precise underlying dynamics remain elusive. In this paper, we argue that post-memorization learning can be understood through the lens of constrained optimization: gradient descent effectively minimizes the weight norm on the zero-loss manifold. We formally prove this in the limit of infinitesimally small le"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2511.01938","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/2511.01938/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":"2511.01938","created_at":"2026-06-02T01:03:37.873228+00:00"},{"alias_kind":"arxiv_version","alias_value":"2511.01938v3","created_at":"2026-06-02T01:03:37.873228+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2511.01938","created_at":"2026-06-02T01:03:37.873228+00:00"},{"alias_kind":"pith_short_12","alias_value":"JR7N5D5K3HXQ","created_at":"2026-06-02T01:03:37.873228+00:00"},{"alias_kind":"pith_short_16","alias_value":"JR7N5D5K3HXQYSCU","created_at":"2026-06-02T01:03:37.873228+00:00"},{"alias_kind":"pith_short_8","alias_value":"JR7N5D5K","created_at":"2026-06-02T01:03:37.873228+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.20441","citing_title":"Weight Decay Regimes in Grokking Transformers: Cheap Online Diagnostics","ref_index":21,"is_internal_anchor":true},{"citing_arxiv_id":"2605.18845","citing_title":"First-Passage Prediction of Grokking Delay: ACalibrated Law under AdamW with Causal Validation","ref_index":2,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JR7N5D5K3HXQYSCUBMNG3CZINO","json":"https://pith.science/pith/JR7N5D5K3HXQYSCUBMNG3CZINO.json","graph_json":"https://pith.science/api/pith-number/JR7N5D5K3HXQYSCUBMNG3CZINO/graph.json","events_json":"https://pith.science/api/pith-number/JR7N5D5K3HXQYSCUBMNG3CZINO/events.json","paper":"https://pith.science/paper/JR7N5D5K"},"agent_actions":{"view_html":"https://pith.science/pith/JR7N5D5K3HXQYSCUBMNG3CZINO","download_json":"https://pith.science/pith/JR7N5D5K3HXQYSCUBMNG3CZINO.json","view_paper":"https://pith.science/paper/JR7N5D5K","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2511.01938&json=true","fetch_graph":"https://pith.science/api/pith-number/JR7N5D5K3HXQYSCUBMNG3CZINO/graph.json","fetch_events":"https://pith.science/api/pith-number/JR7N5D5K3HXQYSCUBMNG3CZINO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JR7N5D5K3HXQYSCUBMNG3CZINO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JR7N5D5K3HXQYSCUBMNG3CZINO/action/storage_attestation","attest_author":"https://pith.science/pith/JR7N5D5K3HXQYSCUBMNG3CZINO/action/author_attestation","sign_citation":"https://pith.science/pith/JR7N5D5K3HXQYSCUBMNG3CZINO/action/citation_signature","submit_replication":"https://pith.science/pith/JR7N5D5K3HXQYSCUBMNG3CZINO/action/replication_record"}},"created_at":"2026-06-02T01:03:37.873228+00:00","updated_at":"2026-06-02T01:03:37.873228+00:00"}