{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:WD4ODE53GDN2V2WSGUWBZLLHMA","short_pith_number":"pith:WD4ODE53","schema_version":"1.0","canonical_sha256":"b0f8e193bb30dbaaead2352c1cad67603edb0fa16b874c7f360d02bc981f47ef","source":{"kind":"arxiv","id":"2602.21160","version":3},"attestation_state":"computed","paper":{"title":"Not Just How Much, But Where: Decomposing Epistemic Uncertainty into Per-Class Contributions","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.AP","stat.ME"],"primary_cat":"stat.ML","authors_text":"David A. Stephens, Mame Diarra Toure","submitted_at":"2026-02-24T18:05:51Z","abstract_excerpt":"In safety-critical classification, the cost of failure is often asymmetric, yet Bayesian deep learning summarises epistemic uncertainty with a single scalar, mutual information (MI), that cannot distinguish whether a model's ignorance involves a benign or safety-critical class. We decompose MI into a per-class vector $C_k(x)=\\sigma_k^{2}/(2\\mu_k)$, with $\\mu_k{=}\\mathbb{E}[p_k]$ and $\\sigma_k^2{=}\\mathrm{Var}[p_k]$ across posterior samples. The decomposition follows from a second-order Taylor expansion of the entropy; the $1/\\mu_k$ weighting corrects boundary suppression and makes $C_k$ compar"},"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":"2602.21160","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"stat.ML","submitted_at":"2026-02-24T18:05:51Z","cross_cats_sorted":["cs.LG","stat.AP","stat.ME"],"title_canon_sha256":"8be68d0359729c77742266aedc7b5914ce31fc7956e5fb41d39a1c608b73605e","abstract_canon_sha256":"24eee2cfa48ab709b4ebd03ce9092f688b98eb38d9605cb38b4a0bc9a0269d9a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-19T16:11:21.156856Z","signature_b64":"Llvacn32jhTsKwDfowQVcbYV3R1igpcd7ZKYOt9wcaOLqMO8lWwRZrLI67BFCgdarAB9CPw0MFp78Q1hlPB7BA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b0f8e193bb30dbaaead2352c1cad67603edb0fa16b874c7f360d02bc981f47ef","last_reissued_at":"2026-06-19T16:11:21.156392Z","signature_status":"signed_v1","first_computed_at":"2026-06-19T16:11:21.156392Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Not Just How Much, But Where: Decomposing Epistemic Uncertainty into Per-Class Contributions","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG","stat.AP","stat.ME"],"primary_cat":"stat.ML","authors_text":"David A. Stephens, Mame Diarra Toure","submitted_at":"2026-02-24T18:05:51Z","abstract_excerpt":"In safety-critical classification, the cost of failure is often asymmetric, yet Bayesian deep learning summarises epistemic uncertainty with a single scalar, mutual information (MI), that cannot distinguish whether a model's ignorance involves a benign or safety-critical class. We decompose MI into a per-class vector $C_k(x)=\\sigma_k^{2}/(2\\mu_k)$, with $\\mu_k{=}\\mathbb{E}[p_k]$ and $\\sigma_k^2{=}\\mathrm{Var}[p_k]$ across posterior samples. The decomposition follows from a second-order Taylor expansion of the entropy; the $1/\\mu_k$ weighting corrects boundary suppression and makes $C_k$ compar"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.21160","kind":"arxiv","version":3},"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/2602.21160/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":"2602.21160","created_at":"2026-06-19T16:11:21.156449+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.21160v3","created_at":"2026-06-19T16:11:21.156449+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.21160","created_at":"2026-06-19T16:11:21.156449+00:00"},{"alias_kind":"pith_short_12","alias_value":"WD4ODE53GDN2","created_at":"2026-06-19T16:11:21.156449+00:00"},{"alias_kind":"pith_short_16","alias_value":"WD4ODE53GDN2V2WS","created_at":"2026-06-19T16:11:21.156449+00:00"},{"alias_kind":"pith_short_8","alias_value":"WD4ODE53","created_at":"2026-06-19T16:11:21.156449+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.21256","citing_title":"Reliable Automated Triage in Spanish Clinical Notes: A Hybrid Framework for Risk-Aware HIV Suspicion Identification","ref_index":37,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WD4ODE53GDN2V2WSGUWBZLLHMA","json":"https://pith.science/pith/WD4ODE53GDN2V2WSGUWBZLLHMA.json","graph_json":"https://pith.science/api/pith-number/WD4ODE53GDN2V2WSGUWBZLLHMA/graph.json","events_json":"https://pith.science/api/pith-number/WD4ODE53GDN2V2WSGUWBZLLHMA/events.json","paper":"https://pith.science/paper/WD4ODE53"},"agent_actions":{"view_html":"https://pith.science/pith/WD4ODE53GDN2V2WSGUWBZLLHMA","download_json":"https://pith.science/pith/WD4ODE53GDN2V2WSGUWBZLLHMA.json","view_paper":"https://pith.science/paper/WD4ODE53","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.21160&json=true","fetch_graph":"https://pith.science/api/pith-number/WD4ODE53GDN2V2WSGUWBZLLHMA/graph.json","fetch_events":"https://pith.science/api/pith-number/WD4ODE53GDN2V2WSGUWBZLLHMA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WD4ODE53GDN2V2WSGUWBZLLHMA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WD4ODE53GDN2V2WSGUWBZLLHMA/action/storage_attestation","attest_author":"https://pith.science/pith/WD4ODE53GDN2V2WSGUWBZLLHMA/action/author_attestation","sign_citation":"https://pith.science/pith/WD4ODE53GDN2V2WSGUWBZLLHMA/action/citation_signature","submit_replication":"https://pith.science/pith/WD4ODE53GDN2V2WSGUWBZLLHMA/action/replication_record"}},"created_at":"2026-06-19T16:11:21.156449+00:00","updated_at":"2026-06-19T16:11:21.156449+00:00"}