Bounded Rational Decision-Making with Adaptive Neural Network Priors