Table 2 Influence of label smoothing and pseudo-labelling on the accuracy (%) in DigitFive
From: A framework for self-supervised federated domain adaptation
Methods | Target | |||||
|---|---|---|---|---|---|---|
Mnist | Mnist-m | Svhn | Syn | Usps | Avg | |
No label smoothing; No pseudo-labelling. (learning rate is 0.05) | ||||||
 SFDA-LS-PL | 99.19 | 74.53 | 83.13 | 89.42 | 97.74 | 88.80 |
No label smoothing. (learning rate is 0.01) | ||||||
 SFDA-LS | 99.26 | 74.09 | 84.80 | 89.29 | 97.69 | 89.03 |
Standard settings. (learning rate is 0.01) | ||||||
 SFDA | 99.10 | 72.31 | 86.02 | 90.37 | 98.06 | 89.17 |