I aim to develop provably reliable and efficient ML systems by combining deep probabilistic models with symbolic reasoning.