Generalized Conditional Gradient and Learning in Potential Mean Field Games