Achieving Zero Constraint Violation for Constrained Reinforcement Learning via Primal-Dual Approach.
Qinbo BaiAmrit Singh BediMridul AgarwalAlec KoppelVaneet AggarwalPublished in: AAAI (2022)
Keyphrases
- primal dual
- saddle point
- reinforcement learning
- linear programming
- convex optimization
- interior point methods
- linear program
- affine scaling
- convergence rate
- linear programming problems
- approximation algorithms
- constraint violations
- hard constraints
- interior point algorithm
- convex constraints
- variational inequalities
- interior point
- semidefinite programming
- inequality constraints
- constrained problems
- algorithm for linear programming
- simplex algorithm
- machine learning
- duality gap
- simplex method
- lagrange multipliers
- convex optimization problems
- dynamic programming
- convex programming
- markov decision processes
- dual formulation
- penalty function
- quadratic programming
- global constraints
- learning algorithm
- special case
- optimal policy
- model free