Approximating High-Dimensional Minimal Surfaces with Physics-Informed Neural Networks