Training neuromorphic hardware to perform specific tasks requires learning algorithms that are robust to a variety of uncertainties and noise sources. My goal is to use theoretical ideas from machine learning and statistics to design algorithms that can effectively learn on neuromorphic hardware.