Sufficient Conditions for Instability of the Subgradient Method with Constant Step Size

DOI10.1137/22M1535723arXiv2211.14852OpenAlexW4390586700MaRDI QIDQ6136655FDOQ6136655

Authors: Cédric Josz, Lexiao Lai 

Publication date: 17 January 2024

Published in: SIAM Journal on Optimization (Search for Journal in Brave)

Abstract: We provide sufficient conditions for instability of the subgradient method with constant step size around a local minimum of a locally Lipschitz semi-algebraic function. They are satisfied by several spurious local minima arising in robust principal component analysis and neural networks.

Full work available at URL: https://arxiv.org/abs/2211.14852

zbMATH Keywords

Lyapunov stabilitysubgradient methodmetric subregularityVerdier condition

Mathematics Subject Classification ID

Numerical mathematical programming methods (65K05) Nonlinear programming (90C30)

Cites Work

• Variational Analysis
• Robust principal component analysis?
• Title not available (Why is that?)
• Clarke Subgradients of Stratifiable Functions
• Local differentiability of distance functions
• Convergence of the Iterates of Descent Methods for Analytic Cost Functions
• Title not available (Why is that?)
• Tangents to an analytic variety
• Characterization of metric regularity of subdifferentials
• Hölder metric subregularity with applications to proximal point method
• Nonconvergence to unstable points in urn models and stochastic approximations
• Metric subregularity of multifunctions: first and second order infinitesimal characterizations
• Higher-order metric subregularity and its applications
• Regularity and conditioning of solution mappings in variational analysis
• Stratifications de Whitney et théorème de Bertini-Sard
• Verdier and strict Thom stratifications in o-minimal structures
• (In-)stability of differential inclusions. Notions, equivalences, and Lyapunov-like characterizations
• On the complexity of robust PCA and \(\ell_1\)-norm low-rank matrix approximation
• Hölder stable minimizers, tilt stability, and Hölder metric regularity of subdifferentials
• Gradient descent only converges to minimizers: non-isolated critical points and invariant regions
• Conservative set valued fields, automatic differentiation, stochastic gradient methods and deep learning
• An Introduction to Optimization on Smooth Manifolds
• Lyapunov stability of the subgradient method with constant step size

This page was built for publication: Sufficient Conditions for Instability of the Subgradient Method with Constant Step Size

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6136655)