PATROL: Provable Defense against Adversarial Policy in Two-player Games

Keyphrases

• Two-player Games 100%
• Policy Training 100%
• Deep Reinforcement Learning (deep RL) 100%
• Adversarial Policies 100%
• Training Methods 66%
• Nash Equilibrium Point 66%
• Robust Policy 66%
• Recent Advances 33%
• Attacker 33%
• Artificial Intelligence 33%
• Individual Choice 33%
• Game Theory 33%
• Sequential Decision Making 33%
• Stock Market 33%
• Competitive Game 33%
• Deep Neural Network 33%
• Training Problems 33%
• Threat Model 33%
• Trading Decisions 33%
• Adversarial Agents 33%
• Adversarial Setting 33%
• Classical Games 33%
• Adversarial Robustness 33%
• Trained Agent 33%
• Agent Policy 33%
• Reinforcement Learning Task 33%
• Policy Space 33%

Computer Science

• Deep Reinforcement Learning 100%
• Training Policy 100%
• Nash Equilibrium 66%
• Equilibrium Point 66%
• Attackers 33%
• Performance Bound 33%
• Sequential Decision Making 33%
• Competitive Game 33%
• Deep Neural Network 33%
• Threat Model 33%
• Individual Decision 33%
• Artificial Intelligence 33%
• Adversarial Setting 33%