{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:7WXDY6METDSIF2HLDIYUQHTTDZ","short_pith_number":"pith:7WXDY6ME","schema_version":"1.0","canonical_sha256":"fdae3c798498e482e8eb1a31481e731e583839d4587d060dbadf8ee3e4d31f78","source":{"kind":"arxiv","id":"2407.12332","version":1},"attestation_state":"computed","paper":{"title":"Why Do You Grok? A Theoretical Analysis of Grokking Modular Addition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Danica J. Sutherland, Lei Wu, Mohamad Amin Mohamadi, Zhiyuan Li","submitted_at":"2024-07-17T06:15:30Z","abstract_excerpt":"We present a theoretical explanation of the ``grokking'' phenomenon, where a model generalizes long after overfitting,for the originally-studied problem of modular addition. First, we show that early in gradient descent, when the ``kernel regime'' approximately holds, no permutation-equivariant model can achieve small population error on modular addition unless it sees at least a constant fraction of all possible data points. Eventually, however, models escape the kernel regime. We show that two-layer quadratic networks that achieve zero training loss with bounded $\\ell_{\\infty}$ norm generali"},"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":"2407.12332","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-07-17T06:15:30Z","cross_cats_sorted":["stat.ML"],"title_canon_sha256":"ce53d89f308c52112a6e76df45ca91ff08f7f8602a486853339720d6d8c33ded","abstract_canon_sha256":"bd3f286e3c11d5898e820be62ab2aaf792a3e086e0b618b2ede58b016d8fee80"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:45:03.020698Z","signature_b64":"2mDYspa+AXgQ+E1m3ogGJN8xVKzlX0d0G20bRlkdsKSvkhjTZEKmxWq6Q5YmNBhLWT8CYNRqphp01buvhzImCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fdae3c798498e482e8eb1a31481e731e583839d4587d060dbadf8ee3e4d31f78","last_reissued_at":"2026-07-05T08:45:03.020215Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:45:03.020215Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Why Do You Grok? A Theoretical Analysis of Grokking Modular Addition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Danica J. Sutherland, Lei Wu, Mohamad Amin Mohamadi, Zhiyuan Li","submitted_at":"2024-07-17T06:15:30Z","abstract_excerpt":"We present a theoretical explanation of the ``grokking'' phenomenon, where a model generalizes long after overfitting,for the originally-studied problem of modular addition. First, we show that early in gradient descent, when the ``kernel regime'' approximately holds, no permutation-equivariant model can achieve small population error on modular addition unless it sees at least a constant fraction of all possible data points. Eventually, however, models escape the kernel regime. We show that two-layer quadratic networks that achieve zero training loss with bounded $\\ell_{\\infty}$ norm generali"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.12332","kind":"arxiv","version":1},"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/2407.12332/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":"2407.12332","created_at":"2026-07-05T08:45:03.020273+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.12332v1","created_at":"2026-07-05T08:45:03.020273+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.12332","created_at":"2026-07-05T08:45:03.020273+00:00"},{"alias_kind":"pith_short_12","alias_value":"7WXDY6METDSI","created_at":"2026-07-05T08:45:03.020273+00:00"},{"alias_kind":"pith_short_16","alias_value":"7WXDY6METDSIF2HL","created_at":"2026-07-05T08:45:03.020273+00:00"},{"alias_kind":"pith_short_8","alias_value":"7WXDY6ME","created_at":"2026-07-05T08:45:03.020273+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.17767","citing_title":"Feature Learning in Linear-Width Two-Layer Networks: Two vs. One Step of Gradient Descent","ref_index":203,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17767","citing_title":"Feature Learning in Linear-Width Two-Layer Networks: Two vs. One Step of Gradient Descent","ref_index":203,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06352","citing_title":"Topological Signatures of Grokking","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7WXDY6METDSIF2HLDIYUQHTTDZ","json":"https://pith.science/pith/7WXDY6METDSIF2HLDIYUQHTTDZ.json","graph_json":"https://pith.science/api/pith-number/7WXDY6METDSIF2HLDIYUQHTTDZ/graph.json","events_json":"https://pith.science/api/pith-number/7WXDY6METDSIF2HLDIYUQHTTDZ/events.json","paper":"https://pith.science/paper/7WXDY6ME"},"agent_actions":{"view_html":"https://pith.science/pith/7WXDY6METDSIF2HLDIYUQHTTDZ","download_json":"https://pith.science/pith/7WXDY6METDSIF2HLDIYUQHTTDZ.json","view_paper":"https://pith.science/paper/7WXDY6ME","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.12332&json=true","fetch_graph":"https://pith.science/api/pith-number/7WXDY6METDSIF2HLDIYUQHTTDZ/graph.json","fetch_events":"https://pith.science/api/pith-number/7WXDY6METDSIF2HLDIYUQHTTDZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7WXDY6METDSIF2HLDIYUQHTTDZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7WXDY6METDSIF2HLDIYUQHTTDZ/action/storage_attestation","attest_author":"https://pith.science/pith/7WXDY6METDSIF2HLDIYUQHTTDZ/action/author_attestation","sign_citation":"https://pith.science/pith/7WXDY6METDSIF2HLDIYUQHTTDZ/action/citation_signature","submit_replication":"https://pith.science/pith/7WXDY6METDSIF2HLDIYUQHTTDZ/action/replication_record"}},"created_at":"2026-07-05T08:45:03.020273+00:00","updated_at":"2026-07-05T08:45:03.020273+00:00"}