Pessimistic Minimax Value Iteration: Provably Efficient Equilibrium Learning from Offline Datasets

Keyphrases

• Function Approximation 100%
• Data Dependency 100%
• Value Function 100%
• Efficient Equilibria 100%
• Distributional Shift 100%
• Approximate Nash Equilibrium 100%
• Value Iteration 100%
• Minimax Optimal 100%
• Markov Games 100%
• Uniform Coverage 100%
• Minimax Values 100%
• Equilibrium Learning 100%
• Information Theory 50%
• Sampling Efficiency 50%
• Optimal Policy 50%
• Equilibrium Policies 50%
• Relative Uncertainty 50%
• Behavioral Policy 50%
• Minimax Optimization 50%
• Large State Space 50%
• Zero-sum Markov Game 50%

Computer Science

• Function Approximation 100%
• Nash Equilibrium 100%
• Function Value 100%
• Theoretical Lower Bound 50%
• Large State Space 50%
• Sufficient Condition 50%

Mathematics

• Minimax 100%
• Nash Equilibrium 40%
• Function Value 40%
• Upper Bound 40%
• Approximation Function 40%
• Dependent Data 40%
• Approximates 40%
• Necessary and Sufficient Condition 20%
• Sample Efficiency 20%
• Optimal Policy 20%