Backdoor Attack on Reinforcement Learning on Pacman Discrete (evaluation)
628.7Backdoor Rate (BR)No Poisoning
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| No PoisoningAgent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 628.7 | — | — | 236.5 | |
| DazeTarget Action=No-Op, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 591.8 | 99.4 | 0.4 | 169.3 | |
| SleeperNetsTarget Action=No-Op, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 525.3 | 100 | 0 | 126.3 | |
| Q-InceptTarget Action=No-Op, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 457.1 | 100 | 0 | 87.3 | |
| TrojDRLTarget Action=No-Op, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 427.8 | 98.6 | 0.7 | 56.3 |