AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

Distribution-Free Calibration Guarantees for Histogram Binning without Sample Splitting

Chirag Gupta, Aaditya Ramdas

ICML 2021 · Learning theory and algorithms › Prediction and decision losses · approach: Probabilistic calibration · 6 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Avoid sample splitting when forming uniform-mass bins and estimating bin biasesUniform-mass binning on CREDIT · CREDITvalidity V(0.05): 0.63 -> 0.79Improved
Avoid 50:50 sample splittingUniform-mass binning on CREDIT · CREDITsample size for (0.1, 0.1) marginal calibration: n = 1000 -> n = 500Improved
Use uniform-mass binning instead of fixed-width binning or isotonic regressionCalibration on CREDIT · CREDITconditional calibrationImproved
Average labels within bins rather than scaling-method predictionsUMD versus scaling-binning on CREDIT · CREDITvalidityMixed
Exclude the boundary score's label from bin-bias estimation (ablation)UMD · Distribution-free calibrationconditional calibration bound ε: Original adds 1/floor n/BImproved
Include boundary points in bin-bias estimation (ablation)Modified UMD · Distribution-free calibrationconditional calibration bound ε: + 1/floor n/BHurt
Add infinitesimal random perturbations to scores and bin biases (ablation)UMD · Distribution-free calibrationmarginal calibration bound ε: sqrt(log(2B/α)/(2(floor n/B - 1))) -> sqrt(log(2/α)/(2(floor n/B - 1))) + δImproved
Use 10 bins rather than fewer bins (ablation)UMD · Distribution-free calibrationcalibration bound εMixed
Use randomized UMD rather than non-randomized UMD when scores may have ties (ablation)UMD · Distribution-free calibrationvalidityImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy