List your company
ResearchFormerly WizWand

Sponsored Search Conversion on Criteo Sponsored Search Conversion (val)

4,128.41Validation MSESwitch-Cond-DP
3,857.49325,686.18167,514.879,343.5584Jun 17, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Switch-Cond-DPprivacy budget (epsilon)=∞, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.064,128.41
DPSGDprivacy budget (epsilon)=∞, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.064,296.44
Cond-DPϵ=∞, Evaluation Protocol=linear model2026.064,726.78
RR-on-Binsprivacy budget (epsilon)=4, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.064,769.61
RR-on-Binsϵ=4, Evaluation Protocol=linear model2026.064,769.61
DPSGDϵ=∞, Evaluation Protocol=linear model2026.065,287.43
Cond-DPϵ=4, Evaluation Protocol=linear model2026.065,297.63
Cond-DPϵ=2, Evaluation Protocol=linear model2026.065,335.86
Cond-DPϵ=1, Evaluation Protocol=linear model2026.065,630.08
Switch-Cond-DPprivacy budget (epsilon)=4, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.066,365.98
DPSGDprivacy budget (epsilon)=4, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.066,444.03
Switch-Cond-DPprivacy budget (epsilon)=2, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.066,529.16
DPSGDϵ=4, Evaluation Protocol=linear model2026.066,565.11
DPSGDϵ=2, Evaluation Protocol=linear model2026.066,589.69
DPSGDϵ=1, Evaluation Protocol=linear model2026.066,673.56
DPSGDϵ=0.5, Evaluation Protocol=linear model2026.066,780.06
DPSGDϵ=0.25, Evaluation Protocol=linear model2026.066,842.63
DPSGDprivacy budget (epsilon)=2, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.066,933.28
Switch-Cond-DPprivacy budget (epsilon)=1, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.066,935
Switch-Cond-DPprivacy budget (epsilon)=0.5, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.066,959.93
Switch-Cond-DPprivacy budget (epsilon)=0.25, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.066,975.9
Cond-DPϵ=0.5, Evaluation Protocol=linear model2026.066,998.62
RR-on-Binsprivacy budget (epsilon)=2, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.067,294.93
RR-on-Binsϵ=2, Evaluation Protocol=linear model2026.067,294.93
DPSGDprivacy budget (epsilon)=1, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.067,472.65
DPSGDprivacy budget (epsilon)=0.5, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.068,073.85
DPSGDprivacy budget (epsilon)=0.25, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.068,533.74
Cond-DPϵ=0.25, Evaluation Protocol=linear model2026.069,677.99
RR-on-Binsprivacy budget (epsilon)=1, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.069,744.08
RR-on-Binsϵ=1, Evaluation Protocol=linear model2026.069,744.08
RR-on-Binsprivacy budget (epsilon)=0.5, prediction head architecture=two-layer MLP (128, 64 hidden units), input transformation=linear2026.0610,901.33
RR-on-Binsϵ=0.5, Evaluation Protocol=linear model2026.0610,901.33