AI Paper Atlas

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

Calibrating Predictions to Decisions: A Novel Approach to Multi-Class Calibration

Shengjia Zhao, Michael P. Kim, Roshni Sahoo, Tengyu Ma, Stefano Ermon

NeurIPS 2021 · Learning theory and algorithms › Prediction and decision losses · approach: Probabilistic calibration · 4 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
add decision recalibration after temperature scalingDenseNet-121 · HAM10000decision loss gap (avg and worst)Improved
add decision recalibrationDenseNet-121 · HAM10000decision lossImproved
add decision recalibrationDenseNet-121 · HAM10000top-1 accuracy: 0.40 \pm 0.08Improved
add decision recalibrationDenseNet-121 · HAM10000L2 loss: 0.010 \pm 0.001Improved
add decision recalibrationpretrained ImageNet classifiers · ImageNet-1000loss gapImproved
add decision recalibrationpretrained ImageNet classifiers · ImageNet-1000accuracy: 0.30Improved
add decision recalibrationpretrained ImageNet classifiers · ImageNet-1000L2 error: 0.00173Improved
replace Dirichlet calibration with decision recalibration as number of classes grows to 1000 (ablation)calibrators · ImageNet subsetsloss gapImproved
add more recalibration steps (ablation)decision recalibration · HAM10000loss gapImproved
more classes (10 to 1000) (ablation)decision recalibration · ImageNetloss gap reductionMixed

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