Accuracy can hide how often post-hoc calibration changes a model’s top-1 prediction. CORD is a post-fit adapter that repairs the calibrated probability vector to recover the original top-1 exactly while preserving the calibrated conditional distribution over the remaining classes, attaining zero Top-1 Prediction Change Rate (TPCR) by construction. Across CIFAR-10/100 and ImageNet-1K, CORD also lowers mean ECE, NLL, and Brier relative to the corresponding direct calibrated outputs.</p>\n<p>Paper: <a href=\"https://arxiv.org/abs/2609.01072\" rel=\"nofollow\">https://arxiv.org/abs/2609.01072</a><br>Code: <a href=\"https://github.com/labhai/CORD\" rel=\"nofollow\">https://github.com/labhai/CORD</a></p>\n","updatedAt":"2026-09-04T06:30:36.814Z","author":{"_id":"67039e7443929668458d3618","avatarUrl":"/avatars/081486010c5d27153410f68cc3325e6a.svg","fullname":"DaehwanKim","name":"HYUDHKIM","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8362177610397339},"editors":["HYUDHKIM"],"editorAvatarUrls":["/avatars/081486010c5d27153410f68cc3325e6a.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2609.01072","authors":[{"_id":"6a9a58aa8f7c3b755723967b","user":{"_id":"67039e7443929668458d3618","avatarUrl":"/avatars/081486010c5d27153410f68cc3325e6a.svg","isPro":false,"fullname":"DaehwanKim","user":"HYUDHKIM","type":"user","name":"HYUDHKIM"},"name":"Daehwan Kim","status":"claimed_verified","statusLastChangedAt":"2026-09-04T08:45:04.344Z","hidden":false},{"_id":"6a9a58aa8f7c3b755723967c","name":"Haejun Chung","hidden":false},{"_id":"6a9a58aa8f7c3b755723967d","name":"Ikbeom Jang","hidden":false}],"publishedAt":"2026-09-02T00:00:00.000Z","submittedOnDailyAt":"2026-09-04T00:00:00.000Z","title":"Let Confidence Change, Not the Prediction: Prediction-Preserving Repair for Post-hoc Calibration","submittedOnDailyBy":{"_id":"67039e7443929668458d3618","avatarUrl":"/avatars/081486010c5d27153410f68cc3325e6a.svg","isPro":false,"fullname":"DaehwanKim","user":"HYUDHKIM","type":"user","name":"HYUDHKIM"},"summary":"Post-hoc calibration corrects reported confidence, yet a multiclass calibrator can also change the associated top-1 prediction. Accuracy captures only the net effect of these changes on correctness, not how often predictions change; the Top-1 Prediction Change Rate (TPCR) instead measures this frequency. We propose Calibrator-Output Repair for Top-1 Decision Preservation (CORD), the first post-fit adapter to impose exact prediction preservation by repairing the full calibrated probability vector. From the original and calibrated outputs alone, CORD determines the mass assigned to the original top-1. The calibrated conditional distribution allocates the remaining mass over the other classes, yielding a repaired vector whose own argmax recovers the original prediction. On the calibration split, CORD coordinates the repaired masses to retain the calibrated outputs' mean mass on original predictions whenever attainable. The adapter alters neither the fitted calibrator nor its direct output, fits no additional supervised map, and requires no user- or validation-tuned hyperparameter. Across CIFAR-10/100 and ImageNet-1K, CORD attains zero TPCR by construction and lowers mean ECE, NLL, and Brier relative to the corresponding direct outputs in every dataset; paired gains persist under distribution shift and across calibration-set sizes. CORD thus removes the preservation constraint from calibrator fitting and assigns exact recovery of the original decision to subsequent output repair. Our code is available at https://github.com/labhai/CORD.","upvotes":1,"discussionId":"6a9a58aa8f7c3b755723967e","githubRepo":"https://github.com/labhai/CORD","githubRepoAddedBy":"user","ai_summary":"CORD is a post-fit adapter that repairs calibrated probability vectors to exactly preserve original top-1 predictions while maintaining calibration quality.","ai_keywords":["post-hoc calibration","multiclass calibrator","Top-1 Prediction Change Rate (TPCR)","CORD","calibrated probability vector","argmax","ECE","NLL","Brier score","distribution shift"],"ai_summary_model":"thinkingmachines/Inkling-Small"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"67039e7443929668458d3618","avatarUrl":"/avatars/081486010c5d27153410f68cc3325e6a.svg","isPro":false,"fullname":"DaehwanKim","user":"HYUDHKIM","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"query":{}}">
Let Confidence Change, Not the Prediction: Prediction-Preserving Repair for Post-hoc Calibration
Abstract
CORD is a post-fit adapter that repairs calibrated probability vectors to exactly preserve original top-1 predictions while maintaining calibration quality.
Post-hoc calibration corrects reported confidence, yet a multiclass calibrator can also change the associated top-1 prediction. Accuracy captures only the net effect of these changes on correctness, not how often predictions change; the Top-1 Prediction Change Rate (TPCR) instead measures this frequency. We propose Calibrator-Output Repair for Top-1 Decision Preservation (CORD), the first post-fit adapter to impose exact prediction preservation by repairing the full calibrated probability vector. From the original and calibrated outputs alone, CORD determines the mass assigned to the original top-1. The calibrated conditional distribution allocates the remaining mass over the other classes, yielding a repaired vector whose own argmax recovers the original prediction. On the calibration split, CORD coordinates the repaired masses to retain the calibrated outputs' mean mass on original predictions whenever attainable. The adapter alters neither the fitted calibrator nor its direct output, fits no additional supervised map, and requires no user- or validation-tuned hyperparameter. Across CIFAR-10/100 and ImageNet-1K, CORD attains zero TPCR by construction and lowers mean ECE, NLL, and Brier relative to the corresponding direct outputs in every dataset; paired gains persist under distribution shift and across calibration-set sizes. CORD thus removes the preservation constraint from calibrator fitting and assigns exact recovery of the original decision to subsequent output repair. Our code is available at https://github.com/labhai/CORD.
Community
Accuracy can hide how often post-hoc calibration changes a model’s top-1 prediction. CORD is a post-fit adapter that repairs the calibrated probability vector to recover the original top-1 exactly while preserving the calibrated conditional distribution over the remaining classes, attaining zero Top-1 Prediction Change Rate (TPCR) by construction. Across CIFAR-10/100 and ImageNet-1K, CORD also lowers mean ECE, NLL, and Brier relative to the corresponding direct calibrated outputs.
Paper: https://arxiv.org/abs/2609.01072
Code: https://github.com/labhai/CORD
Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images
Cite arxiv.org/abs/2609.01072 in a model README.md to link it from this page.
Cite arxiv.org/abs/2609.01072 in a dataset README.md to link it from this page.
Cite arxiv.org/abs/2609.01072 in a Space README.md to link it from this page.
Discussion (0)
Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.
Sign in →No comments yet. Sign in and be the first to say something.