Welcome to the homepage of the

Mathematical Optimization Group

Department of Mathematics, University of Tübingen, Germany

Global Convergence of Model Function Based Bregman Proximal Minimization Algorithms

M.C. Mukkamala, J. Fadili and P. Ochs

Lipschitz continuity of the gradient mapping of a continuously differentiable function plays a crucial role in designing various optimization algorithms. However, many functions arising in practical applications such as low rank matrix factorization or deep neural network problems do not have a Lipschitz continuous gradient. This led to the development of a generalized notion known as the L-smad property, which is based on generalized proximity measures called Bregman distances. However, the L-smad property cannot handle nonsmooth functions, for example, simple nonsmooth functions like |x^4-1| and also many practical composite problems are out of scope. We fix this issue by proposing the MAP property, which generalizes the L-smad property and is also valid for a large class of nonconvex nonsmooth composite problems. Based on the proposed MAP property, we propose a globally convergent algorithm called Model BPG, that unifies several existing algorithms. The convergence analysis is based on a new Lyapunov function. We also numerically illustrate the superior performance of Model BPG on standard phase retrieval problems, robust phase retrieval problems, and Poisson linear inverse problems, when compared to a state of the art optimization method that is valid for generic nonconvex nonsmooth optimization problems.
pdf Bibtex arXiv
Latest update: 24.12.2020
M.C. Mukkamala, J. Fadili, P. Ochs:
Global Convergence of Model Function Based Bregman Proximal Minimization Algorithms. [pdf]
Technical Report, ArXiv e-prints, arXiv:2012.13161, 2020.
  title        = {Global Convergence of Model Function Based Bregman Proximal Minimization Algorithms},
  author       = {M.C. Mukkamala and J. Fadili and P. Ochs},
  year         = {2020},
  journal      = {ArXiv e-prints, arXiv:2012.13161},

MOP Group
The author is not
responsible for
the content of
external pages.