Towards Understanding Gradient Approximation in Equality Constrained Deep Declarative Networks