Natural policy gradient primal-dual method for constrained Markov decision processes

Keyphrases

• Primal-dual Algorithm 100%
• Constrained Markov Decision Process 100%
• Natural Policy Gradient 100%
• Parametrization 40%
• Constraint Violation 40%
• Infinite Horizon 40%
• Optimality Gap 40%
• Nonasymptotic Convergence 40%
• Sampling Methods 20%
• Finite Sample 20%
• Convergence Guarantee 20%
• Convergence Property 20%
• Approximation Error 20%
• Computational Results 20%
• Two-sample 20%
• Convergence Rate 20%
• Softmax 20%
• Dimension-free 20%
• Sequential Decision Problems 20%
• Function Approximation 20%
• Policy-based 20%
• Global Convergence 20%
• Dual Variable 20%
• Gradient Ascent 20%
• Total Rewards 20%
• State Action 20%
• Total Utility 20%
• Nonconcave 20%
• Policy Gradient Method 20%
• Action Space 20%
• Sample Complexity 20%
• Subgradient Descent 20%
• Complexity Guarantees 20%
• Nonconvex Constraint Sets 20%

Computer Science

• Markov Decision Process 100%
• Primal-Dual 100%
• Parametrization 50%
• Constraint Violation 50%
• Objective Function 25%
• Convergence Property 25%
• Function Approximation 25%
• Gradient Descent 25%
• Global Convergence 25%
• Constraint Set 25%
• Sequential Decision Making 25%
• Gradient Method 25%
• Process Problem 25%

Mathematics

• Markov Decision Process 100%
• Asymptotics 50%
• Optimality 50%
• Convergence Property 25%
• Approximation Error 25%
• Objective Function 25%
• Dual Variable 25%
• Action Space 25%