So far we have worked with models whose parameters were handed to us. In reality, those numbers—the noise variances, the persistence coefficients—are almost never known. This chapter shows you how to turn the Kalman filter into a learning machine: we feed it data, and it tells us which parameter values make that data most likely.