List your company

Misclassification Detection on CIFAR-10 (test)

100AUPR SuccSoftmax Baseline
97.358498.044298.7399.4158Oct 7, 2016May 16, 2018Dec 23, 2019Jul 31, 2021Mar 9, 2023Oct 15, 2024May 25, 2026

Evaluation results

MethodLinks
Softmax BaselineArchitecture=40-4 wide residual network, Nonlinearity=GELU, Optimizer=SGD with restarts2016.1010093509543580——
Deep EnsemblesM (Ensemble size)=52026.0599.32————————
MoDEX2026.0599.18————————
MoDEXM (Ensemble size)=12026.0599.18————————
F-EDL2026.0599.1————————
DAEDL2026.0599.08————————
R-EDL2026.0598.98————————
Dropout2026.0598.86————————
RED2026.0598.82————————
Re-EDL2026.0598.81————————
EDL2026.0598.74————————
I-EDL2026.0598.72————————
DUQ2026.0597.89————————
NatPN2026.0597.53————————
PostNet2026.0597.46————————
D_alphaProtocol=TBB (Totally Black-Box), Classifier Accuracy=95%2021.06—94—————17.9—
D_alphaProtocol=PBB (Partially Black-Box), Classifier Accuracy=95%2021.06—95.2—————13.9—
D_betaProtocol=TBB (Totally Black-Box), Classifier Accuracy=95%2021.06—68.5—————18.6—
D_betaProtocol=PBB (Partially Black-Box), Classifier Accuracy=95%2021.06—94.8—————13.4—
DAEDL2025.10————————99.08
DAEDLCalibration Protocol=pre-calibration2026.05————————99.08
Dropout2025.10————————97.15
EDL2025.10————————97.86
EDLCalibration Protocol=pre-calibration2026.05————————98.74
ENERGYProtocol=PBB (Partially Black-Box), Classifier Accuracy=95%2021.06—91.1—————34.7—
F-EDL2025.10————————99.1
GEM-CORECalibration Protocol=pre-calibration2026.05————————99.22
GEM-FICalibration Protocol=pre-calibration2026.05————————99.94
GEM-MIXCalibration Protocol=pre-calibration2026.05————————99.43
I-EDL2025.10————————98.72
I-EDLCalibration Protocol=pre-calibration2026.05————————98.71
MHLNBProtocol=TBB (Totally Black-Box), Classifier Accuracy=95%2021.06—92.2—————30.8—
MHLNBProtocol=PBB (Partially Black-Box), Classifier Accuracy=95%2021.06—84.4—————44.6—
ODINProtocol=TBB (Totally Black-Box), Classifier Accuracy=95%2021.06—93.8—————18.2—
ODINProtocol=PBB (Partially Black-Box), Classifier Accuracy=95%2021.06—94.2—————18.4—
R-EDL2025.10————————98.98
R-EDLCalibration Protocol=pre-calibration2026.05————————98.98
Re-EDLCalibration Protocol=pre-calibration2026.05————————98.81
SRProtocol=TBB (Totally Black-Box), Classifier Accuracy=95%2021.06—93.8—————18.2—
TSCalibration Protocol=post-hoc2026.05————————98.89