Differential Privacy Noise Scale Evaluation on CIFAR-100
95,448Noise Scale (L2)Sensitive DP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Sensitive DPBackbone=CNN, Noise Scale (epsilon)=0.42026.01 | 95,448 | 533 | |
| Full DPBackbone=CNN, Noise Scale (epsilon)=0.22026.01 | 77,523 | 6,968 | |
| AdapLDPBackbone=CNN, Noise Scale (epsilon)=0.22026.01 | 72,781 | 6,685 | |
| AdapLDPBackbone=CNN, Noise Scale (epsilon)=0.32026.01 | 56,700 | 6,685 | |
| Full DPBackbone=CNN, Noise Scale (epsilon)=0.32026.01 | 52,652 | 6,968 | |
| Time Varying DPBackbone=CNN, Noise Scale (epsilon)=0.22026.01 | 50,803 | 5,311 | |
| AdapLDPBackbone=CNN, Noise Scale (epsilon)=0.42026.01 | 44,469 | 6,685 | |
| Full DPBackbone=CNN, Noise Scale (epsilon)=0.42026.01 | 42,236 | 6,968 | |
| DPA LDPBackbone=CNN, Noise Scale (epsilon)=0.32026.01 | 37,329 | 4,135 | |
| DPA LDPBackbone=CNN, Noise Scale (epsilon)=0.22026.01 | 35,603 | 4,135 | |
| Time Varying DPBackbone=CNN, Noise Scale (epsilon)=0.32026.01 | 34,114 | 5,311 | |
| DPA LDPBackbone=CNN, Noise Scale (epsilon)=0.42026.01 | 27,668 | 4,135 | |
| Full DPBackbone=CNN, Noise Scale (epsilon)=0.52026.01 | 25,786 | 6,968 | |
| Sensitive DPBackbone=CNN, Noise Scale (epsilon)=0.22026.01 | 23,854 | 533 | |
| Time Varying DPBackbone=CNN, Noise Scale (epsilon)=0.42026.01 | 23,578 | 5,311 | |
| LaDP-FLBackbone=CNN, Noise Scale (epsilon)=0.22026.01 | 22,968 | — | |
| Sensitive DPBackbone=CNN, Noise Scale (epsilon)=0.32026.01 | 21,271 | 533 | |
| LaDP-FLBackbone=CNN, Noise Scale (epsilon)=0.32026.01 | 19,297 | — | |
| AdapLDPBackbone=CNN, Noise Scale (epsilon)=0.52026.01 | 18,574 | 6,685 | |
| Time Varying DPBackbone=CNN, Noise Scale (epsilon)=0.52026.01 | 18,270 | 5,311 | |
| DPA LDPBackbone=CNN, Noise Scale (epsilon)=0.52026.01 | 10,119 | 4,135 | |
| LaDP-FLBackbone=CNN, Noise Scale (epsilon)=0.42026.01 | 9,028 | — | |
| Sensitive DPBackbone=CNN, Noise Scale (epsilon)=0.52026.01 | 8,933 | 533 | |
| LaDP-FLBackbone=CNN, Noise Scale (epsilon)=0.52026.01 | 8,677 | — |