Research Article
Local Epochs Inefficiency Caused by Device Heterogeneity in Federated Learning
Table 11
The mean ranking of dense feature weights.
| Features | P100_Dense | V100_Dense | K40_Dense | MeanRank |
| Batchsize | 10.3196 | 10.6268 | 9.5132 | 10.1532 | Dim_input | 6.0088 | 4.9544 | 3.854 | 4.939066667 | Dim_output | 7.1052 | 4.882 | 5.9032 | 5.963466667 | Opt_SGD | 4.7872 | 7.2696 | 3.9624 | 5.339733333 | Opt_Adadelta | 4.396 | 4.2684 | 2.2148 | 3.6264 | Opt_Adagrad | 5.1912 | 6.3724 | 7.1868 | 6.250133333 | Opt_Momentum | 8.072 | 7.9004 | 3.562 | 6.511466667 | Opt_Adam | 5.4128 | 3.7576 | 5.1504 | 4.7736 | Opt_RMSProp | 7.35 | 2.5372 | 5.1288 | 5.005333333 | Act_relu | 6.0432 | 10.2896 | 9.9068 | 8.746533333 | Act_tanh | 5.6348 | 7.3096 | 11.452 | 8.132133333 | Act_sigmoid | 7.6792 | 7.832 | 10.1656 | 8.558933333 |
|
|