Reinforcement Learning on LunarLander (Wall-clock training time)
0.8Training Time (min)Independent QHD
Evaluation results
| Method | Links | |
|---|---|---|
| Independent QHDClient heterogeneity setup=Heterogeneous, Number of episodes=600, Number of clients=52026.05 | 0.8 | |
| Independent QHDClient heterogeneity setup=Homogeneous, Number of episodes=600, Number of clients=52026.05 | 2.8 | |
| Truncate FedAvg-QHDClient heterogeneity setup=Heterogeneous, Number of episodes=600, Number of clients=52026.05 | 2.9 | |
| FedQHDClient heterogeneity setup=Heterogeneous, Number of episodes=600, Number of clients=52026.05 | 5.9 | |
| Oracle QHD†Client heterogeneity setup=Heterogeneous, Number of episodes=600, Number of clients=52026.05 | 14.8 | |
| Oracle QHD†Client heterogeneity setup=Homogeneous, Number of episodes=600, Number of clients=52026.05 | 15.3 | |
| Truncate FedAvg-QHDClient heterogeneity setup=Homogeneous, Number of episodes=600, Number of clients=52026.05 | 15.5 | |
| FedQHDClient heterogeneity setup=Homogeneous, Number of episodes=600, Number of clients=52026.05 | 15.5 | |
| Oracle DQN†Client heterogeneity setup=Homogeneous, Number of episodes=600, Number of clients=52026.05 | 24.9 | |
| Distillation FedDQNClient heterogeneity setup=Homogeneous, Number of episodes=600, Number of clients=52026.05 | 31.3 | |
| Distillation FedDQNClient heterogeneity setup=Heterogeneous, Number of episodes=600, Number of clients=52026.05 | 35.2 | |
| FedAvg-DQNClient heterogeneity setup=Homogeneous, Number of episodes=600, Number of clients=52026.05 | 36.8 | |
| Oracle DQN†Client heterogeneity setup=Heterogeneous, Number of episodes=600, Number of clients=52026.05 | 130.4 |