New Bregman proximal type algorithms for solving DC optimization problems?

Difference of Convex (DC) optimization problems have objective functions that are differences between two convex functions. Representative ways solving these the proximal DC algorithms, which require part function $L$-smoothness. In this article, we propose Bregman Proximal Algorithm (BPDCA) for large-scale do not possess Instead, it requires has $L$-smooth adaptable property is exploited in gradient algorithms. addition, an accelerated version, with extrapolation (BPDCAe), a new restart scheme. We show global convergence iterates generated by BPDCA(e) to limiting critical point under assumption Kurdyka-{\L}ojasiewicz or subanalyticity and other weaker conditions than those existing methods. applied our algorithms phase retrieval, can be described both as nonconvex problem problem. Numerical experiments showed BPDCAe outperformed proximal-type because formulation allows larger admissible step sizes.