This was part of Machine Learning and Mean-Field Games

Policy gradients algorithms for mean-field control in continuous time

Huyên Pham, University of Paris 6 and CNRS