AI Paper Atlas

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

Sample Margin-Aware Recalibration of Temperature Scaling

Haolan Guo, Linwei Tao, Haoyang Luo, Minjing Dong, Chang Xu

ICML 2026 · Learning theory and algorithms › Prediction and decision losses · approach: Theory of neural network sensitivity · 0 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
train the margin-to-temperature map with Charbonnier-SoftECE instead of NLL (ablation)SMART · ImageNet-1K ViT-B/16ECE (%): 3.62 -> 0.48Improved
train with Charbonnier-SoftECE instead of NLL (ablation)SMART · ImageNet-1K ResNet-50ECE (%): 0.93 -> 0.52Improved
add Charbonnier smoothing to SoftECE (ablation)SMART · ImageNet-1KECE (%): ResNet-50 0.65 -> 0.52; ViT-B/16 0.89 -> 0.48Improved
use logit margin as input instead of entropy, confidence, all logits, max logit or max-mean (ablation)SMART · ImageNet-1KECE (%): ResNet-50 0.85-0.97 -> 0.52Improved
NLL training of the temperature network for longer (80 epochs) (ablation)SMART with NLL objective · ImageNet ViT-B/32ECE: NLL down 15%, ECE up 8% relative to epoch 30Hurt
replace global temperature (TS) with SMART per-sample margin-based temperatureTemperature Scaling · CIFAR-100 ResNet-50ECE (%): 5.61 -> 1.37Improved
apply SMART post-hoc on top of models trained with calibration lossesNLL, Brier, MMCE, LS-0.05, FLSD-53, FL-3 trained models · CIFAR-10 / CIFAR-100ECE (%): CIFAR-100 ResNet-50 MMCE 15.31 -> 0.86Improved
use SMART (post-hoc) instead of training-time UWG/DFL/MDCAVanilla classifiers · ImageNet / SketchECE (%): ImageNet ViT-B/16 UWG 6.29, DFL 7.62, MDCA 10.79 vs SMART 0.48Improved
use Spline calibration under distribution shift on CNNsSpline · ImageNet-Sketch ResNet-50ECE (%): SMART 0.92 vs Spline 9.76Hurt
increase validation set size from 2 to 512 samples (ablation)SMART · ImageNet ViT-B/32ECE (%)Improved
add more validation data to TS (ablation)Temperature Scaling · ImageNet ViT-B/32ECE (%)No clear effect
use SMART instead of PTS for runtimePTS · ImageNet ResNet-50 runtimeruntime (s): 1050.44 -> 23.03Improved

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