List your company
ResearchFormerly WizWand

Backdoor Attack on STL10

100Attack Success Rate (ASR)InkDrop
-2.793623.893250.5877.2668May 23, 2024Sep 12, 2024Jan 3, 2025Apr 26, 2025Aug 16, 2025Dec 7, 2025Mar 30, 2026
Updated 7mo ago

Evaluation results

MethodLinks
InkDropCondensation Method=DM, Condensation Backbone=ConvNet2026.03100——58.05
InkDropCondensation Method=DAM, Condensation Backbone=ConvNet2026.03100——53.5
DoorpingCondensation Method=DAM, Condensation Backbone=ConvNet2026.03100——53.28
RelaxCondensation Method=DAM, Condensation Backbone=ConvNet2026.03100——53.48
RelaxCondensation Method=DM, Condensation Backbone=ConvNet2026.0399.96——59.62
INACTIVEPre-training Dataset=CIFAR10, Trigger=Invisible (Ours)2024.0599.68—74.02—
DRUPEPre-training Dataset=CIFAR10, Trigger=Patch2024.0596.72—74.43—
RelaxCondensation Method=IDM, Condensation Backbone=ConvNet2026.0395.36——65.82
InkDropCondensation Method=IDM, Condensation Backbone=ConvNet2026.0393.33——64.76
DoorpingCondensation Method=IDM, Condensation Backbone=ConvNet2026.0331.44——66.09
BadEncoderPre-training Dataset=CIFAR10, Trigger=CTRL2024.0516.85—75.73—
DoorpingCondensation Method=DM, Condensation Backbone=ConvNet2026.0314.88——57.73
NaiveCondensation Method=DM, Condensation Backbone=ConvNet2026.0310.32——62.1
SimpleCondensation Method=DAM, Condensation Backbone=ConvNet2026.0310.32——53.46
NaiveCondensation Method=IDM, Condensation Backbone=ConvNet2026.0310.24——66.71
SimpleCondensation Method=IDM, Condensation Backbone=ConvNet2026.0310——65.76
BadEncoderPre-training Dataset=CIFAR10, Trigger=WaNet2024.059.78—72.73—
SimpleCondensation Method=DM, Condensation Backbone=ConvNet2026.039.64——59.72
NaiveCondensation Method=DAM, Condensation Backbone=ConvNet2026.038.76——54.9
BadEncoderPre-training Dataset=CIFAR10, Trigger=Ins-Xpro22024.055.91—74.11—
BadEncoderPre-training Dataset=CIFAR10, Trigger=Ins-Kelvin2024.051.16—74.89—
Clean ModelPre-training Dataset=CIFAR10, Trigger=None2024.05—76.14——