{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:JTKV2QPTSCQTBTW47ZPKN2GMJ3","short_pith_number":"pith:JTKV2QPT","schema_version":"1.0","canonical_sha256":"4cd55d41f390a130cedcfe5ea6e8cc4eda32e9eafbc7605e5f9da54b824f8c99","source":{"kind":"arxiv","id":"2106.04972","version":1},"attestation_state":"computed","paper":{"title":"Understanding Softmax Confidence and Uncertainty","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alexandra Brintrup, Jun Zhu, Tim Pearce","submitted_at":"2021-06-09T10:37:29Z","abstract_excerpt":"It is often remarked that neural networks fail to increase their uncertainty when predicting on data far from the training distribution. Yet naively using softmax confidence as a proxy for uncertainty achieves modest success in tasks exclusively testing for this, e.g., out-of-distribution (OOD) detection. This paper investigates this contradiction, identifying two implicit biases that do encourage softmax confidence to correlate with epistemic uncertainty: 1) Approximately optimal decision boundary structure, and 2) Filtering effects of deep networks. It describes why low-dimensional intuition"},"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":"2106.04972","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2021-06-09T10:37:29Z","cross_cats_sorted":["cs.AI","stat.ML"],"title_canon_sha256":"1899762a7aa991d6cc37742abeaa6181404fdd028dfab03ce0950c3f2b160ad0","abstract_canon_sha256":"224e20adf36fc147e7c696d1c26dad0af7d08f8eaf47f09511b15678b1bfb96d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:47:51.548718Z","signature_b64":"n18GN4ubGTfdz/87t7rpXKoq8QvSLZkTWebhFNKRGOUyIUuhGn7tE0h+zGCqgq1xWTjADr3XRiTn35dwguPOCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4cd55d41f390a130cedcfe5ea6e8cc4eda32e9eafbc7605e5f9da54b824f8c99","last_reissued_at":"2026-07-05T02:47:51.548328Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:47:51.548328Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Understanding Softmax Confidence and Uncertainty","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","stat.ML"],"primary_cat":"cs.LG","authors_text":"Alexandra Brintrup, Jun Zhu, Tim Pearce","submitted_at":"2021-06-09T10:37:29Z","abstract_excerpt":"It is often remarked that neural networks fail to increase their uncertainty when predicting on data far from the training distribution. Yet naively using softmax confidence as a proxy for uncertainty achieves modest success in tasks exclusively testing for this, e.g., out-of-distribution (OOD) detection. This paper investigates this contradiction, identifying two implicit biases that do encourage softmax confidence to correlate with epistemic uncertainty: 1) Approximately optimal decision boundary structure, and 2) Filtering effects of deep networks. It describes why low-dimensional intuition"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2106.04972","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/2106.04972/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":"2106.04972","created_at":"2026-07-05T02:47:51.548385+00:00"},{"alias_kind":"arxiv_version","alias_value":"2106.04972v1","created_at":"2026-07-05T02:47:51.548385+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2106.04972","created_at":"2026-07-05T02:47:51.548385+00:00"},{"alias_kind":"pith_short_12","alias_value":"JTKV2QPTSCQT","created_at":"2026-07-05T02:47:51.548385+00:00"},{"alias_kind":"pith_short_16","alias_value":"JTKV2QPTSCQTBTW4","created_at":"2026-07-05T02:47:51.548385+00:00"},{"alias_kind":"pith_short_8","alias_value":"JTKV2QPT","created_at":"2026-07-05T02:47:51.548385+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.08885","citing_title":"Uncertainty-Aware Transformers: Conformal Prediction for Language Models","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08639","citing_title":"VOLTA: The Surprising Ineffectiveness of Auxiliary Losses for Calibrated Deep Learning","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2604.08885","citing_title":"Uncertainty-Aware Transformers: Conformal Prediction for Language Models","ref_index":17,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JTKV2QPTSCQTBTW47ZPKN2GMJ3","json":"https://pith.science/pith/JTKV2QPTSCQTBTW47ZPKN2GMJ3.json","graph_json":"https://pith.science/api/pith-number/JTKV2QPTSCQTBTW47ZPKN2GMJ3/graph.json","events_json":"https://pith.science/api/pith-number/JTKV2QPTSCQTBTW47ZPKN2GMJ3/events.json","paper":"https://pith.science/paper/JTKV2QPT"},"agent_actions":{"view_html":"https://pith.science/pith/JTKV2QPTSCQTBTW47ZPKN2GMJ3","download_json":"https://pith.science/pith/JTKV2QPTSCQTBTW47ZPKN2GMJ3.json","view_paper":"https://pith.science/paper/JTKV2QPT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2106.04972&json=true","fetch_graph":"https://pith.science/api/pith-number/JTKV2QPTSCQTBTW47ZPKN2GMJ3/graph.json","fetch_events":"https://pith.science/api/pith-number/JTKV2QPTSCQTBTW47ZPKN2GMJ3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JTKV2QPTSCQTBTW47ZPKN2GMJ3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JTKV2QPTSCQTBTW47ZPKN2GMJ3/action/storage_attestation","attest_author":"https://pith.science/pith/JTKV2QPTSCQTBTW47ZPKN2GMJ3/action/author_attestation","sign_citation":"https://pith.science/pith/JTKV2QPTSCQTBTW47ZPKN2GMJ3/action/citation_signature","submit_replication":"https://pith.science/pith/JTKV2QPTSCQTBTW47ZPKN2GMJ3/action/replication_record"}},"created_at":"2026-07-05T02:47:51.548385+00:00","updated_at":"2026-07-05T02:47:51.548385+00:00"}