List your company
ResearchFormerly WizWand

Conformal Prediction on MNIST (test)

100CoverageSCP
24.70444.25263.883.348Sep 4, 2025Oct 1, 2025Oct 29, 2025Nov 25, 2025Dec 23, 2025Jan 19, 2026Feb 16, 2026
Updated 5mo ago

Evaluation results

MethodLinks
SCPn (transductive size)=9, alpha=0.1, noise level (1-epsilon)=0.9, Backbone=LeNet52025.09100——3.20453.1923
TSCPn (transductive size)=6, alpha=0.1, noise level (1-epsilon)=0.9, Backbone=LeNet52025.0992——1.4541.3801
Source-Tuned Pseudo-Calibrationsigma=0.72026.0290.3—0.93——
SCPn (transductive size)=3, alpha=0.1, noise level (1-epsilon)=0.9, Backbone=LeNet52025.0990——2.81622.7429
TSCPn (transductive size)=9, alpha=0.1, noise level (1-epsilon)=0.9, Backbone=LeNet52025.0990——1.49611.2871
Target Calibration (Oracle)sigma=0.72026.0289.2—0.91——
APSn (transductive size)=3, alpha=0.1, noise level (1-epsilon)=0.9, Backbone=LeNet52025.0988——2.83142.8313
SCPn (transductive size)=6, alpha=0.1, noise level (1-epsilon)=0.9, Backbone=LeNet52025.0988——3.11613.086
APSn (transductive size)=6, alpha=0.1, noise level (1-epsilon)=0.9, Backbone=LeNet52025.0988——3.10373.1133
Hard Pseudo-Calibrationsigma=0.72026.0287.1—0.88——
APSn (transductive size)=9, alpha=0.1, noise level (1-epsilon)=0.9, Backbone=LeNet52025.0986——3.1773.1669
SCPTransductive size (n)=3, α=0.3, Noise level (1-ϵ)=0.98, Backbone=LeNet52025.09840—-0.0615—
TSCPn (transductive size)=3, alpha=0.1, noise level (1-epsilon)=0.9, Backbone=LeNet52025.0984——1.44451.4641
Source Calibrationsigma=0.72026.0283.7—0.84——
Target Calibration (Oracle)sigma=1.62026.0283.4—3.32——
Target Calibration (Oracle)sigma=22026.0281.3—5.45——
APSTransductive size (n)=9, α=0.3, Noise level (1-ϵ)=0.98, Backbone=LeNet52025.09740.1469—1.9468—
TSCPTransductive size (n)=6, α=0.3, Noise level (1-ϵ)=0.98, Backbone=LeNet52025.09740—-0.0597—
TSCPTransductive size (n)=9, α=0.3, Noise level (1-ϵ)=0.98, Backbone=LeNet52025.09740—0.0942—
Source-Tuned Pseudo-Calibrationsigma=1.62026.0271.8—1.87——
SCPTransductive size (n)=6, α=0.3, Noise level (1-ϵ)=0.98, Backbone=LeNet52025.09700—-0.0477—
APSTransductive size (n)=6, α=0.3, Noise level (1-ϵ)=0.98, Backbone=LeNet52025.09700—0.3496—
SCPTransductive size (n)=9, α=0.3, Noise level (1-ϵ)=0.98, Backbone=LeNet52025.09660—2.0215—
APSTransductive size (n)=3, α=0.3, Noise level (1-ϵ)=0.98, Backbone=LeNet52025.09640—-0.0615—
TSCPTransductive size (n)=3, α=0.3, Noise level (1-ϵ)=0.98, Backbone=LeNet52025.09640—-0.2146—
Hard Pseudo-Calibrationsigma=1.62026.0253.9—0.83——
Source Calibrationsigma=1.62026.0244—0.55——
Source-Tuned Pseudo-Calibrationsigma=22026.0243.7—1.22——
Hard Pseudo-Calibrationsigma=22026.0234.7—0.82——
Source Calibrationsigma=22026.0227.6—0.5——
CEScore function=THR, alpha=0.01, Setting=Centralized2024.05—2.29———
CEScore function=APS, alpha=0.01, Setting=Centralized2024.05—2.5———
ConfTrScore function=THR, alpha=0.01, Setting=Centralized2024.05—6.28———
ConfTrScore function=APS, alpha=0.01, Setting=Centralized2024.05—2.1———
ConfTr_classScore function=THR, alpha=0.01, Setting=Centralized2024.05—2.09———
ConfTr_classScore function=APS, alpha=0.01, Setting=Centralized2024.05—2.13———
DPI boundScore function=THR, alpha=0.01, Setting=Centralized2024.05—2.24———
DPI boundScore function=APS, alpha=0.01, Setting=Centralized2024.05—2.64———
Model-based Fano boundScore function=THR, alpha=0.01, Setting=Centralized2024.05—2.24———
Model-based Fano boundScore function=APS, alpha=0.01, Setting=Centralized2024.05—2.49———
Simple Fano boundScore function=THR, alpha=0.01, Setting=Centralized2024.05—2.09———
Simple Fano boundScore function=APS, alpha=0.01, Setting=Centralized2024.05—2.12———