{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3BU6Q6UFHRZM3ZQUT6NZO74ZFT","short_pith_number":"pith:3BU6Q6UF","schema_version":"1.0","canonical_sha256":"d869e87a853c72cde6149f9b977f992cdb9dcb1c968c92789766e0ce430937ef","source":{"kind":"arxiv","id":"2605.23476","version":1},"attestation_state":"computed","paper":{"title":"Non-normal spectral signatures of instability in neural network training dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","cond-mat.mtrl-sci","math.OC"],"primary_cat":"cs.LG","authors_text":"Souvik Ghosh","submitted_at":"2026-05-22T10:36:48Z","abstract_excerpt":"Training instabilities in deep networks - loss spikes, oscillatory convergence, and gradient pathologies - are empirically prevalent but lack a rigorous operator-theoretic explanation. We show that the linearized update operators for practically used optimizers are generically non-normal: for Adam, non-normality is controlled by the commutator [H, M] between the Hessian and the diagonal adaptive preconditioner, while for SGD with momentum it arises from the augmented state-space structure of the update map. Applying non-normal stability theory to these operators, we derive a conservative pseud"},"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":"2605.23476","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-22T10:36:48Z","cross_cats_sorted":["cond-mat.dis-nn","cond-mat.mtrl-sci","math.OC"],"title_canon_sha256":"89b27b76b7227ef8d7e98d0aabb9d11e5bf7542100f819c0f72745a6ca6c8108","abstract_canon_sha256":"8cdf5ca97983e2ca6476788894616bbb22acf69d898f37304e1880032b86ca9a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-25T02:01:56.838842Z","signature_b64":"hcMZ1IQDwoqjSbTnOD1hFwsuQ1quwLNxUqDUn8g/bYdiOANXuWNgdAyJPfrbipsBaJpexHjaVWpjKsukIBdYCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d869e87a853c72cde6149f9b977f992cdb9dcb1c968c92789766e0ce430937ef","last_reissued_at":"2026-05-25T02:01:56.838138Z","signature_status":"signed_v1","first_computed_at":"2026-05-25T02:01:56.838138Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Non-normal spectral signatures of instability in neural network training dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cond-mat.dis-nn","cond-mat.mtrl-sci","math.OC"],"primary_cat":"cs.LG","authors_text":"Souvik Ghosh","submitted_at":"2026-05-22T10:36:48Z","abstract_excerpt":"Training instabilities in deep networks - loss spikes, oscillatory convergence, and gradient pathologies - are empirically prevalent but lack a rigorous operator-theoretic explanation. We show that the linearized update operators for practically used optimizers are generically non-normal: for Adam, non-normality is controlled by the commutator [H, M] between the Hessian and the diagonal adaptive preconditioner, while for SGD with momentum it arises from the augmented state-space structure of the update map. Applying non-normal stability theory to these operators, we derive a conservative pseud"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.23476","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/2605.23476/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":"2605.23476","created_at":"2026-05-25T02:01:56.838254+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.23476v1","created_at":"2026-05-25T02:01:56.838254+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.23476","created_at":"2026-05-25T02:01:56.838254+00:00"},{"alias_kind":"pith_short_12","alias_value":"3BU6Q6UFHRZM","created_at":"2026-05-25T02:01:56.838254+00:00"},{"alias_kind":"pith_short_16","alias_value":"3BU6Q6UFHRZM3ZQU","created_at":"2026-05-25T02:01:56.838254+00:00"},{"alias_kind":"pith_short_8","alias_value":"3BU6Q6UF","created_at":"2026-05-25T02:01:56.838254+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/3BU6Q6UFHRZM3ZQUT6NZO74ZFT","json":"https://pith.science/pith/3BU6Q6UFHRZM3ZQUT6NZO74ZFT.json","graph_json":"https://pith.science/api/pith-number/3BU6Q6UFHRZM3ZQUT6NZO74ZFT/graph.json","events_json":"https://pith.science/api/pith-number/3BU6Q6UFHRZM3ZQUT6NZO74ZFT/events.json","paper":"https://pith.science/paper/3BU6Q6UF"},"agent_actions":{"view_html":"https://pith.science/pith/3BU6Q6UFHRZM3ZQUT6NZO74ZFT","download_json":"https://pith.science/pith/3BU6Q6UFHRZM3ZQUT6NZO74ZFT.json","view_paper":"https://pith.science/paper/3BU6Q6UF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.23476&json=true","fetch_graph":"https://pith.science/api/pith-number/3BU6Q6UFHRZM3ZQUT6NZO74ZFT/graph.json","fetch_events":"https://pith.science/api/pith-number/3BU6Q6UFHRZM3ZQUT6NZO74ZFT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3BU6Q6UFHRZM3ZQUT6NZO74ZFT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3BU6Q6UFHRZM3ZQUT6NZO74ZFT/action/storage_attestation","attest_author":"https://pith.science/pith/3BU6Q6UFHRZM3ZQUT6NZO74ZFT/action/author_attestation","sign_citation":"https://pith.science/pith/3BU6Q6UFHRZM3ZQUT6NZO74ZFT/action/citation_signature","submit_replication":"https://pith.science/pith/3BU6Q6UFHRZM3ZQUT6NZO74ZFT/action/replication_record"}},"created_at":"2026-05-25T02:01:56.838254+00:00","updated_at":"2026-05-25T02:01:56.838254+00:00"}