Constrained Reinforcement Learning Has Zero Duality Gap.
Santiago PaternainLuiz F. O. ChamonMiguel Calvo-FullanaAlejandro RibeiroPublished in: CoRR (2019)
Keyphrases
- duality gap
- reinforcement learning
- augmented lagrangian
- primal dual
- set covering problem
- inequality constraints
- lagrange multipliers
- machine learning
- optimality conditions
- optimal solution
- dynamic programming
- linear programming
- convex optimization
- constrained optimization
- linear programming problems
- convex programming
- learning algorithm