Adversarial Attack on Pong
21Cumulative RewardNo Attack (Baseline)
Evaluation Results
| Method | Links | |
|---|---|---|
| No Attack (Baseline)Attack Scenario=White-box, Target Policy=A3C2026.02 | 21 | |
| No Attack (Baseline)Attack Scenario=White-box, Target Policy=PPO2026.02 | 21 | |
| No Attack (Baseline)Attack Scenario=Black-box, Target Policy=A3C2026.02 | 21 | |
| No Attack (Baseline)Attack Scenario=Black-box, Target Policy=PPO2026.02 | 21 | |
| No Attack (Baseline)Attack Scenario=White-box, Target Policy=TRPO2026.02 | 20.64 | |
| No Attack (Baseline)Attack Scenario=Black-box, Target Policy=TRPO2026.02 | 20.64 | |
| FGSMAttack Scenario=Black-box, Target Policy=PPO2026.02 | 20.26 | |
| FGSMAttack Scenario=Black-box, Target Policy=A3C2026.02 | 20.25 | |
| SkipAttack Scenario=Black-box, Target Policy=A3C2026.02 | 20.21 | |
| S-TAttack Scenario=Black-box, Target Policy=PPO2026.02 | 20.01 | |
| FGSMAttack Scenario=Black-box, Target Policy=TRPO2026.02 | 19.88 | |
| S-TAttack Scenario=Black-box, Target Policy=TRPO2026.02 | 19.88 | |
| No Attack (Baseline)Attack Scenario=White-box, Target Policy=DQN2026.02 | 19.83 | |
| No Attack (Baseline)Attack Scenario=Black-box, Target Policy=DQN2026.02 | 19.83 | |
| S-TAttack Scenario=Black-box, Target Policy=A3C2026.02 | 19.65 | |
| SkipAttack Scenario=Black-box, Target Policy=PPO2026.02 | 19.18 | |
| FGSMAttack Scenario=Black-box, Target Policy=DQN2026.02 | 18.38 | |
| S-TAttack Scenario=Black-box, Target Policy=DQN2026.02 | 18.12 | |
| SkipAttack Scenario=Black-box, Target Policy=DQN2026.02 | 18.1 | |
| SkipAttack Scenario=Black-box, Target Policy=TRPO2026.02 | 18.02 | |
| EDGEAttack Scenario=Black-box, Target Policy=PPO2026.02 | -5.98 | |
| EDGEAttack Scenario=Black-box, Target Policy=A3C2026.02 | -6.54 | |
| TSGEAttack Scenario=Black-box, Target Policy=A3C2026.02 | -8.24 | |
| EDGEAttack Scenario=Black-box, Target Policy=TRPO2026.02 | -8.98 | |
| PIAAttack Scenario=Black-box, Target Policy=A3C2026.02 | -9.21 | |
| EDGEAttack Scenario=Black-box, Target Policy=DQN2026.02 | -9.22 | |
| PA-ADAttack Scenario=Black-box, Target Policy=A3C2026.02 | -9.31 | |
| PIAAttack Scenario=Black-box, Target Policy=DQN2026.02 | -9.87 | |
| PA-ADAttack Scenario=White-box, Target Policy=TRPO2026.02 | -9.98 | |
| PA-ADAttack Scenario=White-box, Target Policy=PPO2026.02 | -10.08 | |
| PA-ADAttack Scenario=Black-box, Target Policy=DQN2026.02 | -10.31 | |
| AdvRL-GANAttack Scenario=Black-box, Target Policy=PPO2026.02 | -10.46 | |
| TSGEAttack Scenario=Black-box, Target Policy=PPO2026.02 | -10.73 | |
| S-TAttack Scenario=White-box, Target Policy=A3C2026.02 | -10.84 | |
| AdvRL-GANAttack Scenario=White-box, Target Policy=A3C2026.02 | -11.18 | |
| AdvRL-GANAttack Scenario=Black-box, Target Policy=A3C2026.02 | -11.36 | |
| PIAAttack Scenario=Black-box, Target Policy=TRPO2026.02 | -11.49 | |
| TSGEAttack Scenario=Black-box, Target Policy=DQN2026.02 | -11.53 | |
| AdvRL-GANAttack Scenario=White-box, Target Policy=DQN2026.02 | -12.44 | |
| TSGEAttack Scenario=White-box, Target Policy=A3C2026.02 | -12.47 | |
| S-TAttack Scenario=White-box, Target Policy=TRPO2026.02 | -12.77 | |
| PIAAttack Scenario=Black-box, Target Policy=PPO2026.02 | -13.02 | |
| TSGEAttack Scenario=Black-box, Target Policy=TRPO2026.02 | -13.36 | |
| AdvRL-GANAttack Scenario=Black-box, Target Policy=DQN2026.02 | -13.56 | |
| PA-ADAttack Scenario=Black-box, Target Policy=TRPO2026.02 | -13.86 | |
| EDGEAttack Scenario=White-box, Target Policy=A3C2026.02 | -14.33 | |
| S-TAttack Scenario=White-box, Target Policy=PPO2026.02 | -14.66 | |
| EDGEAttack Scenario=White-box, Target Policy=TRPO2026.02 | -14.89 | |
| EDGEAttack Scenario=White-box, Target Policy=PPO2026.02 | -15.13 | |
| TSGEAttack Scenario=White-box, Target Policy=DQN2026.02 | -15.34 | |
| SkipAttack Scenario=White-box, Target Policy=PPO2026.02 | -15.42 | |
| PA-ADAttack Scenario=Black-box, Target Policy=PPO2026.02 | -15.67 | |
| Advantage-based Adversarial Transformer (AAT)Attack Scenario=Black-box, Target Policy=TRPO2026.02 | -16.06 | |
| AdvRL-GANAttack Scenario=Black-box, Target Policy=TRPO2026.02 | -16.36 | |
| TSGEAttack Scenario=White-box, Target Policy=PPO2026.02 | -16.38 | |
| PIAAttack Scenario=White-box, Target Policy=DQN2026.02 | -17.23 | |
| PIAAttack Scenario=White-box, Target Policy=TRPO2026.02 | -17.41 | |
| PIAAttack Scenario=White-box, Target Policy=PPO2026.02 | -17.73 | |
| S-TAttack Scenario=White-box, Target Policy=DQN2026.02 | -17.86 | |
| PIAAttack Scenario=White-box, Target Policy=A3C2026.02 | -18.34 | |
| Advantage-based Adversarial Transformer (AAT)Attack Scenario=Black-box, Target Policy=DQN2026.02 | -18.34 | |
| SkipAttack Scenario=White-box, Target Policy=DQN2026.02 | -18.35 | |
| AdvRL-GANAttack Scenario=White-box, Target Policy=TRPO2026.02 | -18.4 | |
| Advantage-based Adversarial Transformer (AAT)Attack Scenario=Black-box, Target Policy=A3C2026.02 | -18.45 | |
| SkipAttack Scenario=White-box, Target Policy=TRPO2026.02 | -18.79 | |
| Advantage-based Adversarial Transformer (AAT)Attack Scenario=Black-box, Target Policy=PPO2026.02 | -18.98 | |
| SkipAttack Scenario=White-box, Target Policy=A3C2026.02 | -19.36 | |
| EDGEAttack Scenario=White-box, Target Policy=DQN2026.02 | -19.36 | |
| FGSMAttack Scenario=White-box, Target Policy=PPO2026.02 | -19.56 | |
| TSGEAttack Scenario=White-box, Target Policy=TRPO2026.02 | -19.76 | |
| PA-ADAttack Scenario=White-box, Target Policy=A3C2026.02 | -19.8 | |
| Advantage-based Adversarial Transformer (AAT)Attack Scenario=White-box, Target Policy=PPO2026.02 | -19.87 | |
| Advantage-based Adversarial Transformer (AAT)Attack Scenario=White-box, Target Policy=TRPO2026.02 | -19.97 | |
| AdvRL-GANAttack Scenario=White-box, Target Policy=PPO2026.02 | -20.12 | |
| FGSMAttack Scenario=White-box, Target Policy=DQN2026.02 | -20.31 | |
| FGSMAttack Scenario=White-box, Target Policy=A3C2026.02 | -20.35 | |
| FGSMAttack Scenario=White-box, Target Policy=TRPO2026.02 | -20.48 | |
| Advantage-based Adversarial Transformer (AAT)Attack Scenario=White-box, Target Policy=A3C2026.02 | -20.61 | |
| Advantage-based Adversarial Transformer (AAT)Attack Scenario=White-box, Target Policy=DQN2026.02 | -20.96 | |
| PA-ADAttack Scenario=White-box, Target Policy=DQN2026.02 | -21 |