Reward Certification for Policy Smoothed Reinforcement Learning.
Ronghui MuLeandro Soriano MarcolinoYanghao ZhangTianle ZhangXiaowei HuangWenjie RuanPublished in: AAAI (2024)
Keyphrases
- reinforcement learning
- optimal policy
- reward function
- partially observable environments
- total reward
- policy gradient
- average reward
- action selection
- policy search
- reinforcement learning algorithms
- markov decision process
- markov decision processes
- control policy
- function approximation
- partially observable
- state space
- eligibility traces
- markov decision problems
- state action
- policy iteration
- control policies
- inverse reinforcement learning
- reinforcement learning problems
- action space
- actor critic
- policy evaluation
- function approximators
- agent learns
- approximate dynamic programming
- partially observable markov decision processes
- decision problems
- long run
- third party
- finite horizon
- machine learning
- expected reward
- model free
- state dependent
- multi agent
- temporal difference
- reinforcement learning methods
- dynamic programming
- state and action spaces
- rl algorithms
- average cost
- control problems
- partially observable domains
- infinite horizon
- transition model
- transition probabilities
- reward shaping
- reward signal
- agent receives
- continuous state spaces
- learning process