{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:O2WET26N2ASRJSTG766G2B6N5A","short_pith_number":"pith:O2WET26N","schema_version":"1.0","canonical_sha256":"76ac49ebcdd02514ca66ffbc6d07cde83b4420632baf2f55f7f227b72fc337d3","source":{"kind":"arxiv","id":"2508.21495","version":3},"attestation_state":"computed","paper":{"title":"Rethinking Calibration for Early-Exit Neural Networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bartosz W\\'ojcik, Eric Nalisnick, Filip Szatkowski, Grzegorz Choczy\\'nski, Piotr Kubaty","submitted_at":"2025-08-29T10:21:46Z","abstract_excerpt":"Early-exit neural networks (EENNs) accelerate inference by allowing intermediate classifiers to stop computation once predictions are confident enough. Most methods rely on confidence thresholds for exiting, and consequently, improving classifier calibration is widely assumed to improve performance. In this work, we challenge this assumption and show that calibration alone is not sufficient for EENNs to exploit adaptive computation. To address this insufficiency, we introduce Early-Exit Failure Prediction (EEFP), which accounts for both prediction correctness and the cost of further computatio"},"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":"2508.21495","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-29T10:21:46Z","cross_cats_sorted":[],"title_canon_sha256":"448a7182338085eb8b44e52a323081794d21bee77ce3bb1cf20eefa3e5444565","abstract_canon_sha256":"d24b8e0fddadf52de2c18310fe8b558b6cb162d21c6707f6d22dfdb607da7884"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T02:04:42.566835Z","signature_b64":"GuHcqVhLFkHjcfkI4mIGNWU1LCT2taylJnZLhoR9Q9PPncssc01TMK6Fjc2WkiQ9pDQcpQcK1H9RC6ItvugLDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"76ac49ebcdd02514ca66ffbc6d07cde83b4420632baf2f55f7f227b72fc337d3","last_reissued_at":"2026-05-28T02:04:42.566343Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T02:04:42.566343Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Rethinking Calibration for Early-Exit Neural Networks","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bartosz W\\'ojcik, Eric Nalisnick, Filip Szatkowski, Grzegorz Choczy\\'nski, Piotr Kubaty","submitted_at":"2025-08-29T10:21:46Z","abstract_excerpt":"Early-exit neural networks (EENNs) accelerate inference by allowing intermediate classifiers to stop computation once predictions are confident enough. Most methods rely on confidence thresholds for exiting, and consequently, improving classifier calibration is widely assumed to improve performance. In this work, we challenge this assumption and show that calibration alone is not sufficient for EENNs to exploit adaptive computation. To address this insufficiency, we introduce Early-Exit Failure Prediction (EEFP), which accounts for both prediction correctness and the cost of further computatio"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.21495","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/2508.21495/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":"2508.21495","created_at":"2026-05-28T02:04:42.566404+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.21495v3","created_at":"2026-05-28T02:04:42.566404+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.21495","created_at":"2026-05-28T02:04:42.566404+00:00"},{"alias_kind":"pith_short_12","alias_value":"O2WET26N2ASR","created_at":"2026-05-28T02:04:42.566404+00:00"},{"alias_kind":"pith_short_16","alias_value":"O2WET26N2ASRJSTG","created_at":"2026-05-28T02:04:42.566404+00:00"},{"alias_kind":"pith_short_8","alias_value":"O2WET26N","created_at":"2026-05-28T02:04:42.566404+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/O2WET26N2ASRJSTG766G2B6N5A","json":"https://pith.science/pith/O2WET26N2ASRJSTG766G2B6N5A.json","graph_json":"https://pith.science/api/pith-number/O2WET26N2ASRJSTG766G2B6N5A/graph.json","events_json":"https://pith.science/api/pith-number/O2WET26N2ASRJSTG766G2B6N5A/events.json","paper":"https://pith.science/paper/O2WET26N"},"agent_actions":{"view_html":"https://pith.science/pith/O2WET26N2ASRJSTG766G2B6N5A","download_json":"https://pith.science/pith/O2WET26N2ASRJSTG766G2B6N5A.json","view_paper":"https://pith.science/paper/O2WET26N","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.21495&json=true","fetch_graph":"https://pith.science/api/pith-number/O2WET26N2ASRJSTG766G2B6N5A/graph.json","fetch_events":"https://pith.science/api/pith-number/O2WET26N2ASRJSTG766G2B6N5A/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/O2WET26N2ASRJSTG766G2B6N5A/action/timestamp_anchor","attest_storage":"https://pith.science/pith/O2WET26N2ASRJSTG766G2B6N5A/action/storage_attestation","attest_author":"https://pith.science/pith/O2WET26N2ASRJSTG766G2B6N5A/action/author_attestation","sign_citation":"https://pith.science/pith/O2WET26N2ASRJSTG766G2B6N5A/action/citation_signature","submit_replication":"https://pith.science/pith/O2WET26N2ASRJSTG766G2B6N5A/action/replication_record"}},"created_at":"2026-05-28T02:04:42.566404+00:00","updated_at":"2026-05-28T02:04:42.566404+00:00"}