Image Classification on CIFAR-10 (test) (Provable Robustness Metrics)
79.44AccuracyLipConvnet (+ CRC)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LipConvnet (+ CRC)LipConvnet depth (n)=15, Method variant=+ CRC2022.11 | 79.44 | 66.99 | 52.56 | 38.3 | 1.66 | 4.24 | |
| LipConvnet (+ CRC)LipConvnet depth (n)=25, Method variant=+ CRC2022.11 | 79.19 | 66.28 | 51.74 | 37.99 | 2.01 | 3.82 | |
| LipConvnet (+ CRC)LipConvnet depth (n)=20, Method variant=+ CRC2022.11 | 79.13 | 66.45 | 52.45 | 38.12 | 1.63 | 3.14 | |
| LipConvnet (+ CRC)LipConvnet depth (n)=30, Method variant=+ CRC2022.11 | 78.64 | 66.05 | 51.31 | 37.3 | 4.21 | 6.4 | |
| LipConvnet (+ CRC)LipConvnet depth (n)=35, Method variant=+ CRC2022.11 | 78.57 | 65.94 | 52.04 | 37.63 | 5.84 | 8.76 | |
| LipConvnet (+ CRC)LipConvnet depth (n)=40, Method variant=+ CRC2022.11 | 78.41 | 65.51 | 51.32 | 37.3 | 7.08 | 9.77 | |
| LipConvnet (Baseline)LipConvnet depth (n)=15, Method variant=Baseline2022.11 | 77.78 | 62.75 | 46.34 | 31.38 | — | — | |
| LipConvnet (+ Fast)LipConvnet depth (n)=15, Method variant=+ Fast2022.11 | 77.75 | 62.52 | 46.23 | 31.19 | -0.03 | -0.23 | |
| LipConvnet (Baseline)LipConvnet depth (n)=20, Method variant=Baseline2022.11 | 77.5 | 63.31 | 46.42 | 31.53 | — | — | |
| LipConvnet (Baseline)LipConvnet depth (n)=25, Method variant=Baseline2022.11 | 77.18 | 62.46 | 45.78 | 31.16 | — | — | |
| LipConvnet (+ Fast)LipConvnet depth (n)=20, Method variant=+ Fast2022.11 | 77.13 | 62.05 | 45.86 | 31.13 | -0.37 | -1.26 | |
| LipConvnet (+ Fast)LipConvnet depth (n)=25, Method variant=+ Fast2022.11 | 76.94 | 61.91 | 45.59 | 30.69 | -0.24 | -0.55 | |
| LipConvnet (+ Fast)LipConvnet depth (n)=30, Method variant=+ Fast2022.11 | 74.69 | 58.84 | 43.33 | 28.93 | 0.26 | -0.81 | |
| LipConvnet (Baseline)LipConvnet depth (n)=30, Method variant=Baseline2022.11 | 74.43 | 59.65 | 43.76 | 29.16 | — | — | |
| LipConvnet (+ Fast)LipConvnet depth (n)=35, Method variant=+ Fast2022.11 | 72.91 | 57.58 | 41.52 | 27.37 | 0.18 | 0.4 | |
| LipConvnet (Baseline)LipConvnet depth (n)=35, Method variant=Baseline2022.11 | 72.73 | 57.18 | 42.08 | 28.09 | — | — | |
| LipConvnet (+ Fast)LipConvnet depth (n)=40, Method variant=+ Fast2022.11 | 71.6 | 56.15 | 39.82 | 25.63 | 0.27 | 0.41 | |
| LipConvnet (Baseline)LipConvnet depth (n)=40, Method variant=Baseline2022.11 | 71.33 | 55.74 | 39.32 | 26.06 | — | — |