Cross-Entropy Estimators for Sequential Experiment Design with Reinforcement Learning