LGMLOct 16, 2018

Maximizing Monotone DR-submodular Continuous Functions by Derivative-free Optimization

arXiv:1810.06833v2
Originality Highly original
AI Analysis

This work addresses optimization challenges in machine learning and operations research where gradient information is unavailable or noisy, offering a robust alternative for practitioners.

The paper tackles the problem of maximizing monotone DR-submodular continuous functions without using gradient information, proposing a derivative-free algorithm called LDGM that achieves a (1-e^{-β}-ε)-approximation guarantee, matching the best gradient-based methods, and shows robustness under noise in empirical tests on budget allocation.

In this paper, we study the problem of monotone (weakly) DR-submodular continuous maximization. While previous methods require the gradient information of the objective function, we propose a derivative-free algorithm LDGM for the first time. We define $β$ and $α$ to characterize how close a function is to continuous DR-submodulr and submodular, respectively. Under a convex polytope constraint, we prove that LDGM can achieve a $(1-e^{-β}-ε)$-approximation guarantee after $O(1/ε)$ iterations, which is the same as the best previous gradient-based algorithm. Moreover, in some special cases, a variant of LDGM can achieve a $((α/2)(1-e^{-α})-ε)$-approximation guarantee for (weakly) submodular functions. We also compare LDGM with the gradient-based algorithm Frank-Wolfe under noise, and show that LDGM can be more robust. Empirical results on budget allocation verify the effectiveness of LDGM.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes