We propose a generic computational framework for solving large-scale infinite-horizon, discrete-time dynamic incentive problems with persistent hidden types. First, we combine set-valued dynamic programming techniques with unsupervised machine learning to determine irregularly-shaped feasible...