Contextual Pre-Planning on Reward Machine Abstractions for Enhanced Transfer in Deep Reinforcement Learning.
Guy AzranMohamad H. DaneshStefano V. AlbrechtSarah KerenPublished in: CoRR (2023)
Keyphrases
- reinforcement learning
- transfer learning
- action selection
- macro actions
- function approximation
- state space
- partially observable
- contextual information
- reward shaping
- high level
- optimal policy
- model free
- complex domains
- temporal difference
- knowledge transfer
- planning problems
- dynamic programming
- stochastic domains
- deterministic domains
- markov decision problems
- heuristic search
- multi agent
- temporal abstractions
- average reward
- machine learning
- partially observable environments
- learning algorithm
- reward function
- reinforcement learning methods
- markov decision processes
- partial observability
- batch processing
- reinforcement learning algorithms
- planning process
- function approximators
- partially observable markov decision processes
- control policy
- single agent
- previously learned
- decision theoretic
- context sensitive
- domain independent
- supervised learning