Phase retrieval refers to the reconstruction of a complex-valued vector given the elementwise modulus of its linear measurements. Among the existing modalities, ptychography is particularly popular for its ability to provide high-resolution images by taking overlapping measurements. Nonetheless, the reconstruction of ptychography is typically hindered by an ill-posed loss landscape and lacks convergence guarantees to the global optima. This project aims to explore and improve the convergence and efficiency of the optimization algorithms for ptychography. The student will be provided with codes containing phase retrieval implementations. Various optimization strategies such as quasi-Newton and conjugate gradient will be implemented and compared, either by evaluating the gradient explicitly or through automatic differentiation. Besides, it is also possible to investigate acceleration schedules with GPU and adaptive hyperparameter selection such as linear search. Basic knowledge of Python and optimization is required and experience in PyTorch is recommended.