 In this paper, the authors proposed a novel parametric cost function for model predictive control, MPC, which preserves convexity and allows for the use of sensitivity-based solvers. The parametric cost function was based on a class of neural networks, NNs, and was used to solve economic MPC problems with generic cost functions. The authors showed that their approach could achieve optimal closed-loop performance without requiring an accurate model. This article was authored by Katrina Seal, Arash Bari Kordabad, Sebastian Groh, and others.