neural ordinary differential equations (neural odes)

An approach where the derivatives of the hidden state are parameterized as a neural network.