Physics Informed Neural Networks (PINN) are neural networks, where a library of variables and derivatives are believed to form a PDE which describes the data. The system then seeks a neural network which arrives at a system that replicates a data set.
I have been reading the works of Raissi Mazier regarding PINN and the many other varied approaches he has taken. It is mainly found on Arxiv, so google "raissi arxiv" to find the appropriate downloads.