Updated preprint! “A Simultaneous Approach for Training Neural Differential-Algebraic Systems of Equations” trains neural DAEs, a class of hybrid models that merges physics and machine learning, all at once via nonlinear programming. Check it out on arXiv.