neural ode (ordinary differential equation)

This is a recent advance where the layers of a deep network are defined by a differential equation, allowing the number of layers to become a continuous parameter.