On Centralized and Distributed Mirror Descent: Convergence Analysis Using Quadratic Constraints
نویسندگان
چکیده
Mirror descent (MD) is a powerful first-order optimization technique that subsumes several algorithms including gradient (GD). In this work, we leverage quadratic constraints and Lyapunov functions to analyze the stability characterize convergence rate of MD algorithm as well its distributed variant using semidefinite programming (SDP). For both algorithms, consider strongly convex nonstrongly assumptions. centralized problems, construct an SDP certifies exponential rates derive closed-form feasible solution recovers optimal GD special case. We complement our analysis by providing explicit $O(1/k)$ for problems. Next, numerically whose dimensions are independent network size. To best knowledge, numerical has not been previously reported in literature. further prove setting. Our experiments on problems indicate framework superior compared existing GD.
منابع مشابه
Convergence of Online Mirror Descent Algorithms
In this paper we consider online mirror descent (OMD) algorithms, a class of scalable online learning algorithms exploiting data geometric structures through mirror maps. Necessary and sufficient conditions are presented in terms of the step size sequence {ηt}t for the convergence of an OMD algorithm with respect to the expected Bregman distance induced by the mirror map. The condition is limt→...
متن کاملDistributed Mirror Descent over Directed Graphs
In this paper, we propose Distributed Mirror Descent (DMD) algorithm for constrained convex optimization problems on a (strongly-)connected multi-agent network. We assume that each agent has a private objective function and a constraint set. The proposed DMD algorithm employs a locally designed Bregman distance function at each agent, and thus can be viewed as a generalization of the well-known...
متن کاملConvergence Analysis of Distributed Stochastic Gradient Descent with Shuffling
When using stochastic gradient descent (SGD) to solve large-scale machine learning problems, a common practice of data processing is to shuffle the training data, partition the data across multiple threads/machines if needed, and then perform several epochs of training on the re-shuffled (either locally or globally) data. The above procedure makes the instances used to compute the gradients no ...
متن کاملData-Distributed Weighted Majority and Online Mirror Descent
In this paper, we focus on the question of the extent to which online learning can benefit from distributed computing. We focus on the setting in whichN agents online-learn cooperatively, where each agent only has access to its own data. We propose a generic datadistributed online learning meta-algorithm. We then introduce the Distributed Weighted Majority and Distributed Online Mirror Descent ...
متن کاملGlobal Convergence of Steepest Descent for Quadratic Functions
This paper analyzes the effect of momentum on steepest descent training for quadratic performance functions. Some global convergence conditions of the steepest descent algorithm are obtained by directly analyzing the exact momentum equations for quadratic cost functions. Those conditions can be directly derived from the parameters (different from eigenvalues that are used in the existed ones.) ...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: IEEE Transactions on Automatic Control
سال: 2023
ISSN: ['0018-9286', '1558-2523', '2334-3303']
DOI: https://doi.org/10.1109/tac.2022.3230767