English

Block majorization-minimization with diminishing radius for constrained nonsmooth nonconvex optimization

Optimization and Control 2025-01-22 v6 Machine Learning

Abstract

Block majorization-minimization (BMM) is a simple iterative algorithm for constrained nonconvex optimization that sequentially minimizes majorizing surrogates of the objective function in each block while the others are held fixed. BMM entails a large class of optimization algorithms such as block coordinate descent and its proximal-point variant, expectation-minimization, and block projected gradient descent. We first establish that for general constrained nonsmooth nonconvex optimization, BMM with ρ\rho-strongly convex and LgL_g-smooth surrogates can produce an ϵ\epsilon-approximate first-order optimal point within O~((1+Lg+ρ1)ϵ2)\widetilde{O}((1+L_g+\rho^{-1})\epsilon^{-2}) iterations and asymptotically converges to the set of first-order optimal points. Next, we show that BMM combined with trust-region methods with diminishing radius has an improved complexity of O~((1+Lg)ϵ2)\widetilde{O}((1+L_g) \epsilon^{-2}), independent of the inverse strong convexity parameter ρ1\rho^{-1}, allowing improved theoretical and practical performance with `flat' surrogates. Our results hold robustly even when the convex sub-problems are solved as long as the optimality gaps are summable. Central to our analysis is a novel continuous first-order optimality measure, by which we bound the worst-case sub-optimality in each iteration by the first-order improvement the algorithm makes. We apply our general framework to obtain new results on various algorithms such as the celebrated multiplicative update algorithm for nonnegative matrix factorization by Lee and Seung, regularized nonnegative tensor decomposition, and the classical block projected gradient descent algorithm. Lastly, we numerically demonstrate that the additional use of diminishing radius can improve the convergence rate of BMM in many instances.

Keywords

Cite

@article{arxiv.2012.03503,
  title  = {Block majorization-minimization with diminishing radius for constrained nonsmooth nonconvex optimization},
  author = {Hanbaek Lyu and Yuchen Li},
  journal= {arXiv preprint arXiv:2012.03503},
  year   = {2025}
}

Comments

27 pages, 4 figures. Generalize and improve the analysis for nonsmooth nonconvex problems

R2 v1 2026-06-23T20:46:21.353Z