Skip to Main Content
In this research, we propose a system improving reinforcement learning agents' policies by using a mixture of Bayesian Networks (BNs) to adapt the agents to dynamic environments. A BN is one of stochastic models and used as agents' stochastic knowledge. In our system, models corresponding to new environments are represented by the mixture distribution of BNs constructed in advance.
SICE 2003 Annual Conference (Volume:2 )
Date of Conference: 4-6 Aug. 2003