Reinforcement Learning with Function-Valued Action Spaces for Partial Differential Equation Control.
Yangchen PanAmir-massoud FarahmandMartha WhiteSaleh NabiPiyush GroverDaniel NikovskiPublished in: ICML (2018)
Keyphrases
- action space
- partial differential equations
- reinforcement learning
- function approximators
- control policies
- control policy
- state space
- continuous state
- markov decision processes
- control problems
- state and action spaces
- state action
- action selection
- image denoising
- integral equation
- level set
- real valued
- continuous state spaces
- image processing
- function approximation
- optimal control
- numerical algorithms
- multiscale
- image enhancement
- reinforcement learning algorithms
- stochastic processes
- energy functional
- skill learning
- denoising
- dynamic environments
- continuous action
- policy iteration
- control strategy
- optimal policy
- mobile robot
- learning algorithm
- temporal difference
- policy search
- higher order
- signed distance function
- machine learning