D. Pyeatt (USA)
Knowledge Representation, Machine Learning,Reinforcement Learning, Decision Tree
We present a decision tree based approach to function ap proximation in reinforcement learning. We compare our approach with table lookup and a neural network function approximator on three problems: the well known mountain car and pole balance problems as well as a simulated auto mobile race car. We find that the decision tree can provide better learning performance than the neural network func tion approximation and can solve large problems that are infeasible using table lookup.
Important Links:
Go Back
