Adapting User Interfaces With Model-Based Reinforcement Learning