Expected loss minimization in Differential Privacy on Multimodal Gaussians for Scalar Queries
91.97Improvement over L1 Lossmulti-Gaussian mixture
Evaluation results
| Method | Links | |
|---|---|---|
| multi-Gaussian mixtureepsilon (ε)=10, delta (δ)=0.252026.05 | 91.97 | |
| multi-Gaussian mixtureepsilon (ε)=5, delta (δ)=5 · 10−72026.05 | 85.53 | |
| multi-Gaussian mixtureepsilon (ε)=10, delta (δ)=5 · 10−72026.05 | 82.54 | |
| multi-Gaussian mixtureepsilon (ε)=4, delta (δ)=5 · 10−72026.05 | 79.75 | |
| multi-Gaussian mixtureepsilon (ε)=3, delta (δ)=5 · 10−72026.05 | 78.55 | |
| multi-Gaussian mixtureepsilon (ε)=2, delta (δ)=5 · 10−72026.05 | 65.19 | |
| multi-Gaussian mixtureepsilon (ε)=0.75, delta (δ)=5 · 10−72026.05 | 61.2 | |
| multi-Gaussian mixtureepsilon (ε)=1, delta (δ)=5 · 10−72026.05 | 60.39 | |
| multi-Gaussian mixtureepsilon (ε)=0.5, delta (δ)=5 · 10−72026.05 | 16.74 | |
| multi-Gaussian mixtureepsilon (ε)=0.1, delta (δ)=0.012026.05 | 13.91 | |
| multi-Gaussian mixtureepsilon (ε)=0.1, delta (δ)=0.022026.05 | 8.35 | |
| multi-Gaussian mixtureepsilon (ε)=0.1, delta (δ)=0.052026.05 | 8.05 | |
| multi-Gaussian mixtureepsilon (ε)=0.1, delta (δ)=0.12026.05 | 8.05 | |
| multi-Gaussian mixtureepsilon (ε)=0.1, delta (δ)=0.152026.05 | 3.17 | |
| multi-Gaussian mixtureepsilon (ε)=0.1, delta (δ)=0.252026.05 | 2.19 | |
| multi-Gaussian mixtureepsilon (ε)=0.25, delta (δ)=5 · 10−72026.05 | 2.12 | |
| multi-Gaussian mixtureepsilon (ε)=0.1, delta (δ)=5 · 10−32026.05 | 0.25 | |
| multi-Gaussian mixtureepsilon (ε)=0.1, delta (δ)=10−42026.05 | 0.08 |