Backdoor Attack on Reinforcement Learning on Q*bert Discrete (evaluation)
18,052BRDaze
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DazeTarget Action=Move Right, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 18,052 | 0.993 | 0.2 | 883 | |
| Q-InceptTarget Action=Move Right, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 17,749 | 1 | 0 | 1,380 | |
| No PoisoningAgent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 17,322 | — | — | 1,773 | |
| SleeperNetsTarget Action=Move Right, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 17,194 | 1 | 0 | 1,676 | |
| TrojDRLTarget Action=Move Right, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 17,113 | 0.882 | 6.3 | 1,465 |