Preprint reports improved quantum state preparation through learned gate controls

Researchers report a method that can improve the trade-off between quantum state preparation accuracy and the number of control phases required, according to an arXiv preprint. The method learns which phases a quantum gate needs alongside their values, rather than assigning controls everywhere or restricting them to a predefined pattern. A penalty encourages the removal of unnecessary phases, while a gradual adjustment turns the learned selections into binary decisions. Tests on Fourier and Haar-random target states produced competitive or improved accuracy-resource trade-offs against fully parameterized and predefined sparse approaches. The advantage depends on the target state and the dimension of the quantum system.