Backdoor Attack on Reinforcement Learning on Frogger Discrete (evaluation)
479.8Baseline PerformanceDaze
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DazeTarget Action=Move Down, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 479.8 | 99.9 | 3.1 | 88.1 | |
| SleeperNetsTarget Action=Move Down, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 476.6 | 100 | 0 | 105 | |
| Q-InceptTarget Action=Move Down, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 437.9 | 99.2 | 1.16 | 10.4 | |
| No PoisoningAgent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 380.4 | — | — | 89.1 | |
| TrojDRLTarget Action=Move Down, Agent Architecture=PPO, Action Space=Discrete, Trajectories=100, Training Seeds=52026.02 | 373.2 | 95.7 | 3.1 | 13.3 |