Image Classification on GTSRB Non-IID (Dirichlet alpha = 0.8)
76.5AccuracyFLEX-MoE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| FLEX-MoEIndicator=Loss-driven, Client count C=20, Expert count E=8, Expert selection range kc=[2, 6]2025.12 | 76.5 | 0.0085 | 5,410 | |
| GreedyIndicator=Acc-driven, Client count C=20, Expert count E=8, Expert selection range kc=[2, 6]2025.12 | 68.03 | 0.1786 | 100,093 | |
| GreedyIndicator=Loss-driven, Client count C=20, Expert count E=8, Expert selection range kc=[2, 6]2025.12 | 67.94 | 0.0153 | 9,210 | |
| FLEX-MoEIndicator=Acc-driven, Client count C=20, Expert count E=8, Expert selection range kc=[2, 6]2025.12 | 59.34 | 0.0164 | 6,120 | |
| RandomIndicator=None, Client count C=20, Expert count E=8, Expert selection range kc=[2, 6]2025.12 | 47.31 | 0.0433 | 27,746 |