Convergence and convergence rate of stochastic gradient search in the case of multiple and non-isolated extrema
From MaRDI portal
(Redirected from Publication:2018557)
Abstract: The asymptotic behavior of stochastic gradient algorithms is studied. Relying on results from differential geometry (Lojasiewicz gradient inequality), the single limit-point convergence of the algorithm iterates is demonstrated and relatively tight bounds on the convergence rate are derived. In sharp contrast to the existing asymptotic results, the new results presented here allow the objective function to have multiple and non-isolated minima. The new results also offer new insights into the asymptotic properties of several classes of recursive algorithms which are routinely used in engineering, statistics, machine learning and operations research.
Given a continuously differentiable function \(f=f (\theta)\), \(\theta \in \mathbb{R}^n\), the problem is to determine a solution \(\hat{\theta}\) of the equation \(\nabla f (\theta)=0\). Having an estimate of the gradient \(\nabla f (\theta)\), for the iterative computation of a zero of the gradient, a standard stochastic gradient procedure is considered. Under certain assumptions on the additive noise term and the function \(f\), results on the almost sure convergence and convergence rate are provided. The results are illustrated by means of examples from principle component analysis and maximum likelihood estimation.
Recommendations
- Convergence rates for the stochastic gradient descent method for non-convex objective functions
- Convergence Analysis of Stochastic Algorithms
- The convergence of stochastic gradient algorithms applied to learning in neural networks
- Asymptotic bias of stochastic gradient search
- scientific article; zbMATH DE number 1341059
Cites work
- Analyticity, Convergence, and Convergence Rate of Recursive Maximum-Likelihood Estimation in Hidden Markov Models
- Applications of a Kushner and Clark lemma to general classes of stochastic algorithms
- Convergence of the Iterates of Descent Methods for Analytic Cost Functions
- Criterion algorithms of stochastic optimization
- Estimation and annealing for Gibbsian fields
- Gradient Convergence in Gradient methods with Errors
- scientific article; zbMATH DE number 1817650 (Why is no real title available?)
- scientific article; zbMATH DE number 1818892 (Why is no real title available?)
- scientific article; zbMATH DE number 3875113 (Why is no real title available?)
- scientific article; zbMATH DE number 5348356 (Why is no real title available?)
- scientific article; zbMATH DE number 3723610 (Why is no real title available?)
- scientific article; zbMATH DE number 48727 (Why is no real title available?)
- scientific article; zbMATH DE number 3553601 (Why is no real title available?)
- scientific article; zbMATH DE number 1005357 (Why is no real title available?)
- scientific article; zbMATH DE number 1972910 (Why is no real title available?)
- Introduction to Stochastic Search and Optimization
- Maximum likelihood estimation for spatial models by Markov chain Monte Carlo stochastic approximation
- Nonlinear systems.
- On gradients of functions definable in o-minimal structures
- On semi- and subanalytic geometry
- On the Almost Sure Rate of Convergence of Linear Stochastic Approximation Algorithms
- On the stable equilibrium points of gradient systems
- Semianalytic and subanalytic sets
- Stochastic approximation and its applications
- Sur le problème de la division
- The O.D.E. Method for Convergence of Stochastic Approximation and Reinforcement Learning
- The Łojasiewicz Inequality for Nonsmooth Subanalytic Functions with Applications to Subgradient Dynamical Systems
Cited in
(26)- Optimum parameters and nonasymptotic bounds on the rate of convergence of stochastic algorithms in criterial optimization problems
- Convergence analysis of gradient descent stochastic algorithms
- A numerical approach to optimal coherent quantum LQG controller design using gradient descent
- A sharp convergence rate for a model equation of the asynchronous stochastic gradient descent
- Online inference with multi-modal likelihood functions
- Convergence of constant step stochastic gradient descent for non-smooth non-convex functions
- Almost sure convergence of stochastic gradient processes with matrix step sizes
- Asymptotic optimality in stochastic optimization
- scientific article; zbMATH DE number 3909649 (Why is no real title available?)
- scientific article; zbMATH DE number 3968351 (Why is no real title available?)
- Convergence analysis of smoothed stochastic gradient-type algorithm
- Globally convergent stochastic optimization with optimal asymptotic distribution
- scientific article; zbMATH DE number 1341059 (Why is no real title available?)
- A Globally Convergent Stochastic Approximation
- Convergence Analysis of Stochastic Algorithms
- Convergence rates for the stochastic gradient descent method for non-convex objective functions
- Asymptotic analysis via stochastic differential equations of gradient descent algorithms in statistical and computational paradigms
- Two Time-Scale Stochastic Approximation with Controlled Markov Noise and Off-Policy Temporal-Difference Learning
- Infinite WARM graphs III: strong reinforcement regime
- Urns with multiple drawings and graph-based interaction
- On the existence of minimizers in shallow residual ReLU neural network optimization landscapes
- Convergence of stochastic gradient descent schemes for Łojasiewicz-landscapes
- On the existence of optimal shallow feedforward networks with ReLU activation
- A stochastic use of the Kurdyka-Łojasiewicz property: investigation of optimization algorithms behaviours in a non-convex differentiable framework
- Non-convergence to global minimizers in data driven supervised deep learning: Adam and stochastic gradient descent optimization provably fail to converge to global minimizers in the training of deep neural networks with ReLU activation
- A normal map-based proximal stochastic gradient method: convergence and identification properties
This page was built for publication: Convergence and convergence rate of stochastic gradient search in the case of multiple and non-isolated extrema
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2018557)
