Optimality and Approximation with Policy Gradient Methods in Markov Decision Processes

Keyphrases

• Markov Decision Process 100%
• Optimal Policy 100%
• Policy Gradient Methods 100%
• Global Optimum 50%
• Convergence Rate 50%
• State Distribution 50%
• Global Convergence 50%
• Function Approximation 50%
• Natural Policy Gradient 50%
• Policy Classes 50%
• Value Function 25%
• Approximation Error 25%
• Theoretical Convergence 25%
• Convergence Property 25%
• Finite Sample 25%
• Algorithm Design 25%
• Approximate Value Function 25%
• Globally Optimal Solution 25%
• Effective Method 25%
• Computational Approximation 25%
• State Action 25%
• Comprehensive Characterization 25%
• Fast Rates 25%
• Function-based 25%
• Principled Approach 25%
• Agnostic Learning 25%
• Asymptotic Convergence 25%
• Smoothness Assumption 25%
• Action Space 25%
• Dimension-free 25%
• Learning Results 25%
• Reinforcement Learning Problems 25%
• Softmax 25%
• Discounted Markov Decision Processes 25%
• Parametric Policy 25%
• Entropy Regularizer 25%
• Polynomial Convergence Rate 25%
• Inexact Gradient 25%
• Compatibility Function 25%
• Convergence Guarantee 25%
• Exact Gradients 25%
• Large States 25%

Mathematics

• Optimality 100%
• Markov Decision Process 100%
• Optimal Policy 100%
• Global Optimum 50%
• Worst Case 50%
• State Distribution 50%
• Initial State 50%
• Convergence Rate 50%
• Approximation Function 50%
• Polynomial 25%
• Asymptotics 25%
• Main Result 25%
• Approximates 25%
• Convergence Property 25%
• Parametric 25%
• Action Space 25%
• Approximation Error 25%
• Rate of Convergence 25%
• Entropy 25%