Offline Primal-Dual Reinforcement Learning for Linear MDPs.
Germano GabbianelliGergely NeuNneka OkoloMatteo PapiniPublished in: CoRR (2023)
Keyphrases
- primal dual
- reinforcement learning
- markov decision processes
- linear programming
- affine scaling
- linear program
- approximation algorithms
- convex optimization
- interior point methods
- linear programming problems
- interior point algorithm
- convergence rate
- optimal policy
- simplex algorithm
- state space
- function approximation
- algorithm for linear programming
- function approximators
- semidefinite programming
- variational inequalities
- state and action spaces
- markov decision process
- markov decision problems
- interior point
- dynamic programming
- partially observable
- duality gap
- policy search
- semidefinite
- factored markov decision processes
- machine learning
- action space
- learning algorithm
- reinforcement learning algorithms
- reward function
- simplex method
- dual formulation
- feasible solution
- continuous state and action spaces
- multiscale
- lower bound
- multiresolution
- special case
- worst case
- saddle point
- optimal control
- quadratic programming
- average reward
- policy iteration