The process of galaxy formation and evolution involves a number of interesting and complicated physical processes. However, we have only limited information on the real galaxies we would like to compare to our theories: often just the shape, size, and amount of light given off in a few different color filters. This data set behaves like a high-dimensional nonlinear surface, making it an excellent target for machine learning methods. In this talk, I will discuss how we are using simulations of galaxy formation and evolution to train neural networks that can be used to make inferences about real galaxies, using only our necessarily limited set of observational data.