I would say the underlying mathematics is optimization theory, which also has applications in control but was not specifically designed for it (and indeed, the way you approach optimization in control and ML/AI is quite different)