{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:JKFCN56OFHMJOHR23Y5GOL4HIS","short_pith_number":"pith:JKFCN56O","schema_version":"1.0","canonical_sha256":"4a8a26f7ce29d8971e3ade3a672f8744b87dbe2337423e0a435b164753bd5685","source":{"kind":"arxiv","id":"2402.15555","version":2},"attestation_state":"computed","paper":{"title":"Deep Networks Always Grok and Here is Why","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Ahmed Imtiaz Humayun, Randall Balestriero, Richard Baraniuk","submitted_at":"2024-02-23T18:59:31Z","abstract_excerpt":"Grokking, or delayed generalization, is a phenomenon where generalization in a deep neural network (DNN) occurs long after achieving near zero training error. Previous studies have reported the occurrence of grokking in specific controlled settings, such as DNNs initialized with large-norm parameters or transformers trained on algorithmic datasets. We demonstrate that grokking is actually much more widespread and materializes in a wide range of practical settings, such as training of a convolutional neural network (CNN) on CIFAR10 or a Resnet on Imagenette. We introduce the new concept of dela"},"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":"2402.15555","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-23T18:59:31Z","cross_cats_sorted":["cs.AI","cs.CV"],"title_canon_sha256":"42597b6d8466ac0f80a830c5a944d3bf37ba66431f123a53f2dbf70698fb6fb0","abstract_canon_sha256":"6b76148b413204a1d9952c0f0f552af9c1a93e66ed68834d33daeaad46ac0613"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:28:38.878003Z","signature_b64":"VcmdnneCACRplXygTT8dJLDBWNlrmw+1p/XHNNojqJMV6iaz7oRfPcieoGn462Xt+MWrqAnPAuvYMdfLpaoDAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4a8a26f7ce29d8971e3ade3a672f8744b87dbe2337423e0a435b164753bd5685","last_reissued_at":"2026-07-05T08:28:38.877572Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:28:38.877572Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Deep Networks Always Grok and Here is Why","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.CV"],"primary_cat":"cs.LG","authors_text":"Ahmed Imtiaz Humayun, Randall Balestriero, Richard Baraniuk","submitted_at":"2024-02-23T18:59:31Z","abstract_excerpt":"Grokking, or delayed generalization, is a phenomenon where generalization in a deep neural network (DNN) occurs long after achieving near zero training error. Previous studies have reported the occurrence of grokking in specific controlled settings, such as DNNs initialized with large-norm parameters or transformers trained on algorithmic datasets. We demonstrate that grokking is actually much more widespread and materializes in a wide range of practical settings, such as training of a convolutional neural network (CNN) on CIFAR10 or a Resnet on Imagenette. We introduce the new concept of dela"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.15555","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/2402.15555/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":"2402.15555","created_at":"2026-07-05T08:28:38.877635+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.15555v2","created_at":"2026-07-05T08:28:38.877635+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.15555","created_at":"2026-07-05T08:28:38.877635+00:00"},{"alias_kind":"pith_short_12","alias_value":"JKFCN56OFHMJ","created_at":"2026-07-05T08:28:38.877635+00:00"},{"alias_kind":"pith_short_16","alias_value":"JKFCN56OFHMJOHR2","created_at":"2026-07-05T08:28:38.877635+00:00"},{"alias_kind":"pith_short_8","alias_value":"JKFCN56O","created_at":"2026-07-05T08:28:38.877635+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2510.17381","citing_title":"Beyond Binary Out-of-Distribution Detection: Characterizing Distributional Shifts with Multi-Statistic Diffusion Trajectories","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24393","citing_title":"Complexity of Linear Regions in Self-supervised Deep ReLU Networks","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06352","citing_title":"Topological Signatures of Grokking","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08884","citing_title":"HM-Bench: A Comprehensive Benchmark for Multimodal Large Language Models in Hyperspectral Remote Sensing","ref_index":25,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JKFCN56OFHMJOHR23Y5GOL4HIS","json":"https://pith.science/pith/JKFCN56OFHMJOHR23Y5GOL4HIS.json","graph_json":"https://pith.science/api/pith-number/JKFCN56OFHMJOHR23Y5GOL4HIS/graph.json","events_json":"https://pith.science/api/pith-number/JKFCN56OFHMJOHR23Y5GOL4HIS/events.json","paper":"https://pith.science/paper/JKFCN56O"},"agent_actions":{"view_html":"https://pith.science/pith/JKFCN56OFHMJOHR23Y5GOL4HIS","download_json":"https://pith.science/pith/JKFCN56OFHMJOHR23Y5GOL4HIS.json","view_paper":"https://pith.science/paper/JKFCN56O","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.15555&json=true","fetch_graph":"https://pith.science/api/pith-number/JKFCN56OFHMJOHR23Y5GOL4HIS/graph.json","fetch_events":"https://pith.science/api/pith-number/JKFCN56OFHMJOHR23Y5GOL4HIS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JKFCN56OFHMJOHR23Y5GOL4HIS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JKFCN56OFHMJOHR23Y5GOL4HIS/action/storage_attestation","attest_author":"https://pith.science/pith/JKFCN56OFHMJOHR23Y5GOL4HIS/action/author_attestation","sign_citation":"https://pith.science/pith/JKFCN56OFHMJOHR23Y5GOL4HIS/action/citation_signature","submit_replication":"https://pith.science/pith/JKFCN56OFHMJOHR23Y5GOL4HIS/action/replication_record"}},"created_at":"2026-07-05T08:28:38.877635+00:00","updated_at":"2026-07-05T08:28:38.877635+00:00"}