Explore/Exploit Strategies in Autonomy
Stewart W. Wilson · The MIT Press eBooks · 1996
Within a reinforcement learning framework, ten strategies for autonomous control of the explore/exploit decision are reviewed, with observations from initial experiments on four of them. Control based on prediction error or its rate of change appears promising. Connections are made with explore/exploit work by Holland (1975), Thrun (1992), and Schmidhuber (1995a,b).