Accelerated Primal-Dual Policy Optimization for Safe Reinforcement Learning.
Qingkai LiangFanyu QueEytan H. ModianoPublished in: CoRR (2018)
Keyphrases
- primal dual
- reinforcement learning
- saddle point
- linear programming
- optimal policy
- interior point methods
- linear program
- line search
- linear programming problems
- convex optimization
- convex programming
- affine scaling
- convex optimization problems
- approximation algorithms
- convergence rate
- policy search
- algorithm for linear programming
- markov decision process
- variational inequalities
- simplex algorithm
- interior point algorithm
- semidefinite programming
- interior point
- simplex method
- action selection
- quadratic programming
- function approximation
- dynamic programming
- dual formulation
- function approximators
- multiple objectives
- convex functions
- learning algorithm
- state space
- duality gap
- least squares
- optimization problems
- markov decision processes
- model free
- reward function
- reinforcement learning algorithms
- integer programming
- policy gradient
- step size
- solving problems
- semidefinite
- convex relaxation
- machine learning