Image Classification on CIFAR10 (Standard and Lp Robustness Metrics)
96.09Standard AccuracyWang et al. (Trained with ℓ2)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Wang et al. (Trained with ℓ2)Backbone=WRN-70-16, Training Threat Model=ℓ22023.07 | 96.09 | 56.64 | 86.72 | 5.27 | |
| Base WRN-70-16Backbone=WRN-70-162023.07 | 95.31 | — | — | — | |
| ScoreOptBackbone=WRN-70-162023.07 | 95.18 | 71.48 | 87.11 | 93.36 | |
| Wang et al. (Trained with ℓ∞)Backbone=WRN-70-16, Training Threat Model=ℓ∞2023.07 | 92.97 | 72.46 | 71.68 | 3.13 |