Gradient descent with adaptive stepsize converges (nearly) linearly under fourth-order growth
Authors:
Preprint
Abstract
A prevalent belief among optimization specialists is that linear convergence of gradient descent is contingent on the function growing quadratically away from its minimizers. In this work, we argue that this belief is inaccurate. We show that gradient descent with an adaptive stepsize converges at a local (nearly) linear rate on any smooth function that merely exhibits fourth-order growth away from its minimizer. The adaptive stepsize we propose arises from an intriguing decomposition theorem: any such function admits a smooth manifold around the optimal solution -- which we call the ravine -- so that the function grows at least quadratically away from the ravine and has constant order growth along it. The ravine allows one to interlace many short gradient steps with a single long Polyak gradient step, which together ensure rapid convergence to the minimizer. We illustrate the theory and algorithm on the problems of matrix sensing and factorization and learning a single neuron in the overparameterized regime.Keywords
All Related Versions
This publication has 30 references indexed in Scilit:
- Acceleration by stepsize hedging: Silver Stepsize Schedule for smooth convex optimizationMathematical Programming, 2024
- Proximal Methods Avoid Active Strict Saddles of Weakly Convex FunctionsFoundations of Computational Mathematics, 2021
- Low-Rank Matrix Recovery with Composite Optimization: Good Conditioning and Rapid ConvergenceFoundations of Computational Mathematics, 2021
- Implementable tensor methods in unconstrained convex optimizationMathematical Programming, 2019
- Linear convergence of first order methods for non-strongly convex optimizationMathematical Programming, 2018
- Convergence of descent methods for semi-algebraic and tame problems: proximal algorithms, forward–backward splitting, and regularized Gauss–Seidel methodsMathematical Programming, 2011
- Local Minima and Convergence in Low-Rank Semidefinite ProgrammingMathematical Programming, 2004
- A nonlinear programming algorithm for solving semidefinite programs via low-rank factorizationMathematical Programming, 2003
- Error bounds and convergence analysis of feasible descent methods: a general approachAnnals of Operations Research, 1993
- On Richardson's Method for Solving Linear Systems with Positive Definite MatricesJournal of Mathematics and Physics, 1953