Hybrid Reward Architecture for Reinforcement Learning.
Harm van SeijenMehdi FatemiRomain LarocheJoshua RomoffTavian BarnesJeffrey TsangPublished in: NIPS (2017)
Keyphrases
- reinforcement learning
- learning capabilities
- function approximation
- state space
- management system
- real time
- model free
- markov decision processes
- learning agent
- eligibility traces
- hybrid learning
- software architecture
- partially observable
- reinforcement learning algorithms
- optimal control
- architectural design
- learning algorithm
- machine learning
- reinforcement learning methods
- function approximators
- partially observable environments
- design methodology
- reward function
- temporal difference
- dynamic programming
- multi agent systems
- multi agent
- data sets