{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GX6ME2OBOZ4SRA6X2YZXIVMYM7","short_pith_number":"pith:GX6ME2OB","schema_version":"1.0","canonical_sha256":"35fcc269c176792883d7d63374559867c631b4cdb068a1c2bc5d86776585fc7a","source":{"kind":"arxiv","id":"2504.12700","version":1},"attestation_state":"computed","paper":{"title":"A Two-Phase Perspective on Deep Learning Dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","cs.LG"],"primary_cat":"hep-th","authors_text":"Animik Ghosh, Robert de Mello Koch","submitted_at":"2025-04-17T06:57:37Z","abstract_excerpt":"We propose that learning in deep neural networks proceeds in two phases: a rapid curve fitting phase followed by a slower compression or coarse graining phase. This view is supported by the shared temporal structure of three phenomena: grokking, double descent and the information bottleneck, all of which exhibit a delayed onset of generalization well after training error reaches zero. We empirically show that the associated timescales align in two rather different settings. Mutual information between hidden layers and input data emerges as a natural progress measure, complementing circuit-base"},"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":"2504.12700","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"hep-th","submitted_at":"2025-04-17T06:57:37Z","cross_cats_sorted":["cond-mat.dis-nn","cs.LG"],"title_canon_sha256":"dae2ceda4b5116ff6fae5f38ed4cec1132088de1f835204329ce72b94694a507","abstract_canon_sha256":"bf51b2d31229f727e122e9e648fe9911a0ccef50fa588c1cfa0d7f6952969d42"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:43:11.191273Z","signature_b64":"W3iJ/q7N/TjIat3QfsRzB99melxMbYCVxRrzRFBbju5xEc0qK6USXRHar3KX/CmlT1btM0i3khlwI74enPcIBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"35fcc269c176792883d7d63374559867c631b4cdb068a1c2bc5d86776585fc7a","last_reissued_at":"2026-07-05T11:43:11.190783Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:43:11.190783Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A Two-Phase Perspective on Deep Learning Dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","cs.LG"],"primary_cat":"hep-th","authors_text":"Animik Ghosh, Robert de Mello Koch","submitted_at":"2025-04-17T06:57:37Z","abstract_excerpt":"We propose that learning in deep neural networks proceeds in two phases: a rapid curve fitting phase followed by a slower compression or coarse graining phase. This view is supported by the shared temporal structure of three phenomena: grokking, double descent and the information bottleneck, all of which exhibit a delayed onset of generalization well after training error reaches zero. We empirically show that the associated timescales align in two rather different settings. Mutual information between hidden layers and input data emerges as a natural progress measure, complementing circuit-base"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.12700","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/2504.12700/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":"2504.12700","created_at":"2026-07-05T11:43:11.190845+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.12700v1","created_at":"2026-07-05T11:43:11.190845+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.12700","created_at":"2026-07-05T11:43:11.190845+00:00"},{"alias_kind":"pith_short_12","alias_value":"GX6ME2OBOZ4S","created_at":"2026-07-05T11:43:11.190845+00:00"},{"alias_kind":"pith_short_16","alias_value":"GX6ME2OBOZ4SRA6X","created_at":"2026-07-05T11:43:11.190845+00:00"},{"alias_kind":"pith_short_8","alias_value":"GX6ME2OB","created_at":"2026-07-05T11:43:11.190845+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GX6ME2OBOZ4SRA6X2YZXIVMYM7","json":"https://pith.science/pith/GX6ME2OBOZ4SRA6X2YZXIVMYM7.json","graph_json":"https://pith.science/api/pith-number/GX6ME2OBOZ4SRA6X2YZXIVMYM7/graph.json","events_json":"https://pith.science/api/pith-number/GX6ME2OBOZ4SRA6X2YZXIVMYM7/events.json","paper":"https://pith.science/paper/GX6ME2OB"},"agent_actions":{"view_html":"https://pith.science/pith/GX6ME2OBOZ4SRA6X2YZXIVMYM7","download_json":"https://pith.science/pith/GX6ME2OBOZ4SRA6X2YZXIVMYM7.json","view_paper":"https://pith.science/paper/GX6ME2OB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.12700&json=true","fetch_graph":"https://pith.science/api/pith-number/GX6ME2OBOZ4SRA6X2YZXIVMYM7/graph.json","fetch_events":"https://pith.science/api/pith-number/GX6ME2OBOZ4SRA6X2YZXIVMYM7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GX6ME2OBOZ4SRA6X2YZXIVMYM7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GX6ME2OBOZ4SRA6X2YZXIVMYM7/action/storage_attestation","attest_author":"https://pith.science/pith/GX6ME2OBOZ4SRA6X2YZXIVMYM7/action/author_attestation","sign_citation":"https://pith.science/pith/GX6ME2OBOZ4SRA6X2YZXIVMYM7/action/citation_signature","submit_replication":"https://pith.science/pith/GX6ME2OBOZ4SRA6X2YZXIVMYM7/action/replication_record"}},"created_at":"2026-07-05T11:43:11.190845+00:00","updated_at":"2026-07-05T11:43:11.190845+00:00"}