A Self-Supervised Auxiliary Loss for Deep RL in Partially Observable Settings.
Eltayeb AhmedLuisa M. ZintgrafChristian A. Schröder de WittNicolas UsunierPublished in: CoRR (2021)
Keyphrases
- partially observable
- reinforcement learning
- partially observable domains
- markov decision processes
- state space
- partial observability
- markov decision problems
- decision problems
- optimal policy
- function approximation
- partial observations
- dynamical systems
- infinite horizon
- partially observable environments
- hidden state
- reward function
- reinforcement learning algorithms
- fully observable
- machine learning
- partially observable markov decision processes
- markov decision process
- action space
- dynamic programming
- average reward
- action selection
- belief state
- policy iteration
- action models
- planning domains
- transfer learning
- learning algorithm
- model free
- optimal control
- state variables
- function approximators
- long run