Are AlphaZero-like Agents Robust to Adversarial Perturbations?

Keyphrases

• AlphaZero 100%
• Adversarial Perturbation 100%
• Adversarial States 100%
• Neural Network 66%
• Search Space 66%
• Monte Carlo Tree Search 66%
• Adversarial Examples 66%
• Amateur 33%
• Discrete Search 33%
• Human Performance 33%
• Superhuman 33%
• Large Margin 33%
• Original State 33%
• Board Games 33%
• Nogo-A 33%
• Adversarial Attack 33%
• Self-play 33%
• Policy Values 33%
• AlphaGo Zero 33%
• Game of Go 33%
• Meaningless Actions 33%
• Legal State 33%
• Amateur Players 33%
• Differentiable Search 33%

Computer Science

• Neural Network 100%
• Tree Search 100%
• Search Space 100%
• Adversarial Example 100%
• Human Performance 50%
• Adversarial Machine Learning 50%