AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Precision-Induced Miscalibration: Understanding and Correcting Confidence Distortion in Quantized Neural Networks

Jiawei Gu, Fengyuan Nie, Hao Tang, Yanpeng Sun

ICML 2026 · Architectures and efficiency › Quantization · approach: Low-precision training formats

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 PACS sample-adaptive temperature at inferenceResNet-50 FP16 · ImageNet-1K valECE (%): 5.82 -> 1.92Improved
use adaptive PACS temperature instead of fixed global T=1.5 (ablation)PACS · ImageNet-1K val ResNet-50ECE (%): 3.45 -> 1.92Improved
use true precision risk score instead of random risk (ablation)PACS · ImageNet-1K val ResNet-50ECE (%): 4.87 -> 1.92Improved
use L-inf norm for logit scale instead of L2 (ablation)PACS · ImageNet-1K val ResNet-50ECE (%): 2.34 -> 1.92Improved
add stop-gradient on T(x) (ablation)PACS (training) · ImageNet ResNet-50ECE (%) / stability: 2.34 -> 1.92; 3 of 10 runs NaN withoutImproved
add PACS on top of Label SmoothingLabel Smoothing · ImageNet-1K val ResNet-50ECE (%): 3.21 -> 1.54Improved
add PACS on top of Logit NormalizationLN · ImageNet-1K val ResNet-50ECE (%): 2.05 -> 1.75Improved
apply Unit ScalingStandard FP16 · ImageNet-1K val ResNet-50ECE (%): 5.82 -> 5.65No clear effect
apply PACS in INT8 post-training quantizationResNet-50 · ImageNet-1K valECE (%): 5.23 -> 4.56 (vs LN)Improved
apply PACS on FP32 precisionResNet-50 · ImageNet-1K valECE (%): 3.18 vs 3.18 (LN)No clear effect
apply PACS to LLaMA-7B BF16LLaMA-7B · C4ECE: 12.3% -> 7.8%Improved
run trunk in FP16 with FP32 softmax vs fully FP32ResNet-50 · ImageNet-1K valECE degradation (points): +1.68 (trunk), +0.57 (softmax), +2.61 (both)Hurt

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