List your company

Deep Leakage Defense on CIFAR-10

1SSIMGround Truth
0.104560.337030.56950.80197Jan 21, 2026

Evaluation results

MethodLinks
Ground TruthBackbone=ResNet18, Batch size=8, Attack Method=SME2026.0110—03.843173—————
Deep Leakage with Generative Flow Matching DenoiserVRAM(GB)=1.14, TFLOPs=1.1152026.010.6210.66920.861————0.2880.0870.0230.057—
NoneBackbone=ResNet18, Batch size=8, Attack Method=SME2026.010.6091.08319.1270.3233.832173—————
SMEVRAM(GB)=0.824, TFLOPs=0.3022026.010.560.77418.314————0.3120.0750.0310.064—
ClippingBackbone=ResNet18, Batch size=8, Attack Method=SME2026.010.461.33916.7770.5153.1261.0366—————
CompressionBackbone=ResNet18, Batch size=8, Attack Method=SME2026.010.3211.60813.8820.4623.0671.256—————
Gaussian NoiseBackbone=ResNet18, Batch size=8, Attack Method=SME2026.010.2732.21513.5090.5152.3531.0263—————
FedADGBackbone=ResNet18, Batch size=8, Attack Method=SME2026.010.1821.56212.3320.4922.6511.365—————
DLFVRAM(GB)=0.654, TFLOPs=3.662026.010.1711.82712.708————0.6080.2390.0680.332—
GGLVRAM(GB)=2.574, TFLOPs=5.6772026.010.1642.94811.415————0.5050.1140.0920.092—
GradInversionVRAM(GB)=0.742, TFLOPs=0.0032026.010.1590.9928.695————0.5980.180.1520.653—
DCSBackbone=ResNet18, Batch size=8, Attack Method=SME2026.010.1552.64512.2230.5861.8224.1860—————
SpooFLBackbone=ResNet18, Batch size=8, Attack Method=SME2026.010.1396.51311.5070.5761.689163—————
HCBMNEC=5, Aggregation=EBM2026.05———————————0
HCBMNEC=5, Aggregation=TreeHFD2026.05———————————0
HCBMNEC=10, Aggregation=EBM2026.05———————————3
HCBMNEC=10, Aggregation=TreeHFD2026.05———————————0
Linear CBMNEC=52026.05———————————43
Linear CBMNEC=102026.05———————————49