Adaptive sampling using variational autoencoder and reinforcement learning