Projection Methods via Neural Networks for Continuous-Time Models

I extend traditional projection methods by using neural networks as a function approximator to solve continuous-time models. The method is well-suited for high-dimensional settings such as those that arise in the presence of many assets, and can capture the strong non-linearities that occur for instance when agents face constraints. It is designed to accommodate multiple unknown functions, e.g. multiple value functions, so that it is set to tackle economies with several agents and incomplete markets