search

davide97l / rl-policies-attacks-defenses

Adversarial attacks on Deep Reinforcement Learning (RL)
MIT License
76 stars 12 forks source link

A question on 'example.ipynb' #26

Open 2019211753 opened 1 year ago

2019211753 commented 1 year ago

'Attack Pong-PPO policy with Uniform Attack with attack frequenc7 0.5. Moreover, let's suppose we don't know the agent policy is PPO and let's perform attacks on a A2C policy trained on the same environment.' What does it mean?

2019211753 commented 1 year ago

For example, using A2C as the victim policy instead of the targeted PPO policy to initialize the C&W attack, but attacking PPO, what is the effect? Is this to verify the transferability in black-box scenarios?Thanks a lot!