Learning in Sinusoidal Spaces with Physics-Informed Neural Networks