{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:3V7SYW4G2357KRLWH4ESWW6RSL","short_pith_number":"pith:3V7SYW4G","schema_version":"1.0","canonical_sha256":"dd7f2c5b86d6fbf545763f092b5bd192d2f586b272e74911db65d28c0ca6bdcd","source":{"kind":"arxiv","id":"2310.05918","version":1},"attestation_state":"computed","paper":{"title":"Grokking as Compression: A Nonlinear Complexity Perspective","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Max Tegmark, Ziming Liu, Ziqian Zhong","submitted_at":"2023-10-09T17:59:18Z","abstract_excerpt":"We attribute grokking, the phenomenon where generalization is much delayed after memorization, to compression. To do so, we define linear mapping number (LMN) to measure network complexity, which is a generalized version of linear region number for ReLU networks. LMN can nicely characterize neural network compression before generalization. Although the $L_2$ norm has been a popular choice for characterizing model complexity, we argue in favor of LMN for a number of reasons: (1) LMN can be naturally interpreted as information/computation, while $L_2$ cannot. (2) In the compression phase, LMN ha"},"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":"2310.05918","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2023-10-09T17:59:18Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"2203e59062cc1535f4d3267ecfcc0a9297235d26ec39b79980bd19b9c6fd78fb","abstract_canon_sha256":"da0ed278cd2832dc9e9c9b20f08ef9932ef651768471b62d49933fa3e480b0ad"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:58:51.293966Z","signature_b64":"/oXstNjLnegcsZdLup2D67A4SxwLZPFVM+wbb7KLyq5woLGviUD2f3QrZqeIOiLQOI3QTfYtcZBlAV/h+xnZDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dd7f2c5b86d6fbf545763f092b5bd192d2f586b272e74911db65d28c0ca6bdcd","last_reissued_at":"2026-07-05T06:58:51.293485Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:58:51.293485Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Grokking as Compression: A Nonlinear Complexity Perspective","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Max Tegmark, Ziming Liu, Ziqian Zhong","submitted_at":"2023-10-09T17:59:18Z","abstract_excerpt":"We attribute grokking, the phenomenon where generalization is much delayed after memorization, to compression. To do so, we define linear mapping number (LMN) to measure network complexity, which is a generalized version of linear region number for ReLU networks. LMN can nicely characterize neural network compression before generalization. Although the $L_2$ norm has been a popular choice for characterizing model complexity, we argue in favor of LMN for a number of reasons: (1) LMN can be naturally interpreted as information/computation, while $L_2$ cannot. (2) In the compression phase, LMN ha"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2310.05918","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/2310.05918/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":"2310.05918","created_at":"2026-07-05T06:58:51.293543+00:00"},{"alias_kind":"arxiv_version","alias_value":"2310.05918v1","created_at":"2026-07-05T06:58:51.293543+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2310.05918","created_at":"2026-07-05T06:58:51.293543+00:00"},{"alias_kind":"pith_short_12","alias_value":"3V7SYW4G2357","created_at":"2026-07-05T06:58:51.293543+00:00"},{"alias_kind":"pith_short_16","alias_value":"3V7SYW4G2357KRLW","created_at":"2026-07-05T06:58:51.293543+00:00"},{"alias_kind":"pith_short_8","alias_value":"3V7SYW4G","created_at":"2026-07-05T06:58:51.293543+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.16325","citing_title":"Phase Transitions in Driven Informational Systems: A Two-Field Perspective on Learning Theory and Non-Equilibrium Chemistry","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2603.11749","citing_title":"Truth as a Compression Artifact in Language Model Training","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3V7SYW4G2357KRLWH4ESWW6RSL","json":"https://pith.science/pith/3V7SYW4G2357KRLWH4ESWW6RSL.json","graph_json":"https://pith.science/api/pith-number/3V7SYW4G2357KRLWH4ESWW6RSL/graph.json","events_json":"https://pith.science/api/pith-number/3V7SYW4G2357KRLWH4ESWW6RSL/events.json","paper":"https://pith.science/paper/3V7SYW4G"},"agent_actions":{"view_html":"https://pith.science/pith/3V7SYW4G2357KRLWH4ESWW6RSL","download_json":"https://pith.science/pith/3V7SYW4G2357KRLWH4ESWW6RSL.json","view_paper":"https://pith.science/paper/3V7SYW4G","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2310.05918&json=true","fetch_graph":"https://pith.science/api/pith-number/3V7SYW4G2357KRLWH4ESWW6RSL/graph.json","fetch_events":"https://pith.science/api/pith-number/3V7SYW4G2357KRLWH4ESWW6RSL/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3V7SYW4G2357KRLWH4ESWW6RSL/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3V7SYW4G2357KRLWH4ESWW6RSL/action/storage_attestation","attest_author":"https://pith.science/pith/3V7SYW4G2357KRLWH4ESWW6RSL/action/author_attestation","sign_citation":"https://pith.science/pith/3V7SYW4G2357KRLWH4ESWW6RSL/action/citation_signature","submit_replication":"https://pith.science/pith/3V7SYW4G2357KRLWH4ESWW6RSL/action/replication_record"}},"created_at":"2026-07-05T06:58:51.293543+00:00","updated_at":"2026-07-05T06:58:51.293543+00:00"}