On-Line Reinforcement Learning Using Cascade Constructive Neural Networks

In order to scale to problems with large or continuous state-spaces, reinforcement learning algorithms need to use function approximation. Neural networks are one commonly used approach, with most work so far using fixed-architecture networks. Previous supervised learning research has shown that constructive networks which grow their architecture during training outperform fixed-architecture networks. This paper extends the sarsa algorithm to use a cascade constructive network, and shows it outperforms a fixed-architecture network on two benchmark tasks.