Why Do Local Methods Solve Nonconvex Problems?
Abstract
Non-convex optimization is ubiquitous in modern machine learning. Researchers devise non-convex objective functions and optimize them using off-the-shelf optimizers such as stochastic gradient descent and its variants, which leverage the local geometry and update iteratively. Even though solving non-convex functions is NP-hard in the worst case, the optimization quality in practice is often not an issue -- optimizers are largely believed to find approximate global minima. Researchers hypothesize a unified explanation for this intriguing phenomenon: most of the local minima of the practically-used objectives are approximately global minima. We rigorously formalize it for concrete instances of machine learning problems.
Cite
@article{arxiv.2103.13462,
title = {Why Do Local Methods Solve Nonconvex Problems?},
author = {Tengyu Ma},
journal= {arXiv preprint arXiv:2103.13462},
year = {2021}
}
Comments
This is the Chapter 21 of the book "Beyond the Worst-Case Analysis of Algorithms"