Improving Reinforcement Learning Pre-Training with Variational Dropout | IEEE Conference Publication | IEEE Xplore