Research Article
Local Epochs Inefficiency Caused by Device Heterogeneity in Federated Learning
Table 9
The standard deviation of dense feature weight ranking.
| Features | P100_Dense | V100_Dense | K40_Dense | MeanRankStd |
| Batchsize | 1.8927 | 1.4465 | 1.5556 | 1.6316 | Dim_input | 3.3842 | 2.2099 | 2.3296 | 2.6412 | Dim_output | 2.4491 | 2.7088 | 1.7920 | 2.3166 | Opt_SGD | 2.8395 | 2.9443 | 2.1840 | 2.6559 | Opt_Adadelta | 3.3851 | 2.9903 | 1.3891 | 2.5882 | Opt_Adagrad | 2.0481 | 3.0305 | 2.0840 | 2.3875 | Opt_Momentum | 2.8527 | 2.9168 | 1.9885 | 2.5860 | Opt_Adam | 2.7488 | 2.9804 | 2.4706 | 2.7333 | Opt_RMSProp | 3.4788 | 1.7232 | 2.0469 | 2.4163 | Act_relu | 1.8273 | 1.4856 | 1.0920 | 1.4683 | Act_tanh | 2.1289 | 2.2291 | 0.7319 | 1.6966 | Act_sigmoid | 2.2241 | 2.1797 | 1.1457 | 1.8498 |
|
|