Neural Approximations for Optimal Control and Decision