Dynamic Learning of Correlation Potentials for a Time-Dependent Kohn-Sham System