Bandit-Based Policy Invariant Explicit Shaping for Incorporating External Advice in Reinforcement Learning.
Yash SatsangiPaniz BehboudianPublished in: CoRR (2023)
Keyphrases
- reinforcement learning
- optimal policy
- reward shaping
- policy search
- markov decision process
- action selection
- markov decision problems
- reinforcement learning algorithms
- reinforcement learning problems
- markov decision processes
- state and action spaces
- partially observable environments
- partially observable
- actor critic
- action space
- control policy
- function approximation
- function approximators
- reward function
- policy evaluation
- policy gradient
- control policies
- state space
- multi armed bandit
- model free
- policy iteration
- partially observable markov decision processes
- affine invariant
- affine transformation
- approximate dynamic programming
- agent learns
- state dependent
- decision problems
- state action
- partially observable domains
- machine learning
- rl algorithms
- neural network
- average reward
- multi agent
- learning process
- random sampling
- temporal difference
- invariant features
- policy gradient methods
- dynamic programming
- bandit problems
- inverse reinforcement learning
- internal and external
- complex domains
- control problems
- reinforcement learning methods
- continuous state