Sentence Classification on GLUE (MNLI, QQP, QNLI, SST2) (dev)
85.91MNLI-m Accuracyfull AUTO-V
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| full AUTO-VPrivacy budget (ε)=∞ (non-DP), Backbone=RoBERTa-base, Number of parameters=125M2022.06 | 85.91 | 86.14 | 87.34 | 91.4 | 94.49 | |
| full AUTO-SPrivacy budget (ε)=8, Backbone=RoBERTa-base, Number of parameters=125M2022.06 | 83.82 | 83.55 | 86.58 | 87.85 | 92.43 | |
| RGPPrivacy budget (ε)=∞ (non-DP), Backbone=RoBERTa-base, Number of parameters=125M2022.06 | 83.6 | 83.2 | 89.3 | 91.3 | 92.9 | |
| full AUTO-SPrivacy budget (ε)=3, Backbone=RoBERTa-base, Number of parameters=125M2022.06 | 83.22 | 83.21 | 85.76 | 86.91 | 92.32 | |
| fullPrivacy budget (ε)=8, Backbone=RoBERTa-base, Number of parameters=125M2022.06 | 83.2 | 83.46 | 86.08 | 87.94 | 92.09 | |
| fullPrivacy budget (ε)=3, Backbone=RoBERTa-base, Number of parameters=125M2022.06 | 82.45 | 82.99 | 85.56 | 87.42 | 91.86 | |
| full AUTO-VPrivacy budget (ε)=8, Backbone=RoBERTa-base, Number of parameters=125M2022.06 | 82.18 | 82.64 | 86.23 | 87.24 | 92.09 | |
| full AUTO-VPrivacy budget (ε)=3, Backbone=RoBERTa-base, Number of parameters=125M2022.06 | 81.21 | 82.03 | 84.72 | 86.56 | 91.86 | |
| RGPPrivacy budget (ε)=8, Backbone=RoBERTa-base, Number of parameters=125M2022.06 | 80.5 | 79.6 | 85.5 | 87.2 | 91.6 |