Runtime Efficiency (Time-to-Threshold) on MNIST (Image Classification)
89Max AccuracyFAR-SIGN
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FAR-SIGNModel=MLP, Byzantine workers=24%, Clients=512026.05 | 89 | 24 | 51 | |
| Multi-KrumModel=MLP, Byzantine workers=24%, Clients=512026.05 | 87.2 | 249 | 480 | |
| KrumModel=MLP, Byzantine workers=24%, Clients=512026.05 | 85.5 | 331 | — | |
| RFAModel=MLP, Byzantine workers=24%, Clients=512026.05 | 85.4 | 312 | 839 | |
| BulyanModel=MLP, Byzantine workers=24%, Clients=512026.05 | 85.1 | 420 | 885 | |
| MedianModel=MLP, Byzantine workers=24%, Clients=512026.05 | 83.8 | 350 | — | |
| Trimmed MeanModel=MLP, Byzantine workers=24%, Clients=512026.05 | 8.3 | — | — |