An ant system based exploration-exploitation for reinforcement learning
Hyeong Soo Chang · 2004
In this paper, we develop a novel exploration-exploitation strategy for reinforcement learning based on ant colony system. Most of the exploration-exploitation strategies use some statistics extracted from a single simulated trajectory. The novel strategy uses some statistics extracted from multiple simulated trajectories obtained from a swarm of ants. We show that the strategy preserves the convergence property of Q-learning.