{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:BXBJCHFWTPK2VWAEIXEGR3S2ZR","short_pith_number":"pith:BXBJCHFW","schema_version":"1.0","canonical_sha256":"0dc2911cb69bd5aad80445c868ee5acc71cc8ff6591390afb76fd61c35c7456e","source":{"kind":"arxiv","id":"2008.05912","version":2},"attestation_state":"computed","paper":{"title":"A statistical theory of cold posteriors in deep neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Laurence Aitchison","submitted_at":"2020-08-13T13:46:58Z","abstract_excerpt":"To get Bayesian neural networks to perform comparably to standard neural networks it is usually necessary to artificially reduce uncertainty using a \"tempered\" or \"cold\" posterior. This is extremely concerning: if the prior is accurate, Bayes inference/decision theory is optimal, and any artificial changes to the posterior should harm performance. While this suggests that the prior may be at fault, here we argue that in fact, BNNs for image classification use the wrong likelihood. In particular, standard image benchmark datasets such as CIFAR-10 are carefully curated. We develop a generative m"},"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":"2008.05912","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"stat.ML","submitted_at":"2020-08-13T13:46:58Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"c5db0dba308f124b8b683b9d339f6a88ec2cb85d1355a39c9e02c87f8f56adbf","abstract_canon_sha256":"19d2ea96ce6bf8ef2002c4add7d56231a7475a4dee14180425dff2ea200ffb24"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:35:16.633978Z","signature_b64":"rNI5B/2ESanMvAsXElULQ7z5C4zC6oPOg3HZhK6rTY7pyrILMmty3a+1tMfOg3DtFDBemI0J/jdtcy0xl4ZvDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0dc2911cb69bd5aad80445c868ee5acc71cc8ff6591390afb76fd61c35c7456e","last_reissued_at":"2026-07-05T02:35:16.633497Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:35:16.633497Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A statistical theory of cold posteriors in deep neural networks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"stat.ML","authors_text":"Laurence Aitchison","submitted_at":"2020-08-13T13:46:58Z","abstract_excerpt":"To get Bayesian neural networks to perform comparably to standard neural networks it is usually necessary to artificially reduce uncertainty using a \"tempered\" or \"cold\" posterior. This is extremely concerning: if the prior is accurate, Bayes inference/decision theory is optimal, and any artificial changes to the posterior should harm performance. While this suggests that the prior may be at fault, here we argue that in fact, BNNs for image classification use the wrong likelihood. In particular, standard image benchmark datasets such as CIFAR-10 are carefully curated. We develop a generative m"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2008.05912","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/2008.05912/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":"2008.05912","created_at":"2026-07-05T02:35:16.633563+00:00"},{"alias_kind":"arxiv_version","alias_value":"2008.05912v2","created_at":"2026-07-05T02:35:16.633563+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2008.05912","created_at":"2026-07-05T02:35:16.633563+00:00"},{"alias_kind":"pith_short_12","alias_value":"BXBJCHFWTPK2","created_at":"2026-07-05T02:35:16.633563+00:00"},{"alias_kind":"pith_short_16","alias_value":"BXBJCHFWTPK2VWAE","created_at":"2026-07-05T02:35:16.633563+00:00"},{"alias_kind":"pith_short_8","alias_value":"BXBJCHFW","created_at":"2026-07-05T02:35:16.633563+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.25745","citing_title":"Gaussian Mean Field Variational Inference can Overestimate Predictive Variance","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BXBJCHFWTPK2VWAEIXEGR3S2ZR","json":"https://pith.science/pith/BXBJCHFWTPK2VWAEIXEGR3S2ZR.json","graph_json":"https://pith.science/api/pith-number/BXBJCHFWTPK2VWAEIXEGR3S2ZR/graph.json","events_json":"https://pith.science/api/pith-number/BXBJCHFWTPK2VWAEIXEGR3S2ZR/events.json","paper":"https://pith.science/paper/BXBJCHFW"},"agent_actions":{"view_html":"https://pith.science/pith/BXBJCHFWTPK2VWAEIXEGR3S2ZR","download_json":"https://pith.science/pith/BXBJCHFWTPK2VWAEIXEGR3S2ZR.json","view_paper":"https://pith.science/paper/BXBJCHFW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2008.05912&json=true","fetch_graph":"https://pith.science/api/pith-number/BXBJCHFWTPK2VWAEIXEGR3S2ZR/graph.json","fetch_events":"https://pith.science/api/pith-number/BXBJCHFWTPK2VWAEIXEGR3S2ZR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BXBJCHFWTPK2VWAEIXEGR3S2ZR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BXBJCHFWTPK2VWAEIXEGR3S2ZR/action/storage_attestation","attest_author":"https://pith.science/pith/BXBJCHFWTPK2VWAEIXEGR3S2ZR/action/author_attestation","sign_citation":"https://pith.science/pith/BXBJCHFWTPK2VWAEIXEGR3S2ZR/action/citation_signature","submit_replication":"https://pith.science/pith/BXBJCHFWTPK2VWAEIXEGR3S2ZR/action/replication_record"}},"created_at":"2026-07-05T02:35:16.633563+00:00","updated_at":"2026-07-05T02:35:16.633563+00:00"}