检索结果-内蒙古大学图书馆

Avoiding excess computation in asynchronous evolutionary algorithms

EXPERT SYSTEMS 2023年第5期40卷 e13100-e13100页

作者： Scott, Eric O. Coletti, Mark Schuman, Catherine D. Kay, Bill Kulkarni, Shruti R. Parsa, Maryam Gunaratne, Chathika De Jong, Kenneth A. MITRE Corp Mclean VA 22102 USA George Mason Univ Dept Comp Sci Fairfax VA 22030 USA Oak Ridge Natl Lab Computat & Comp Sci Directorate Oak Ridge TN USA Univ Tennessee Dept Elect Engn & Comp Sci Knoxville TN USA Pacific Northwest Natl Lab Algorithms Combinator & Optimizat Richland WA 99352 USA George Mason Univ Dept Elect & Comp Engn Fairfax VA 22030 USA

asynchronous evolutionary algorithms are becoming increasingly popular as a means of making full use of many processors while solving computationally expensive search and optimization problems. These algorithms excel at keeping large clusters fully utilized, but may sometimes inefficiently sample an excess of fast-evaluating solutions at the expense of higher-quality, slow-evaluating ones. We have previously introduced a steady-state parent selection strategy, SWEET ("Selection whilE EvaluaTing"), that sometimes selects individuals that are still being evaluated and allows them to reproduce early. We perform a takeover-time analysis that confirms that this strategy gives slow-evaluating individuals that have higher fitnesses an increased ability to multiply in the population. We also find that SWEET appears effective at improving optimization performance on problems in which solution quality is positively correlated with evaluation time. We evaluate our approach on six simulated real-valued optimization problems and three real-world applications: an autonomous vehicle controller problem that involves tuning a spiking neural network and two adversarial EA problems. We further evaluate SWEET versus a basic asynchronous process in a simulated setting. We present evidence that SWEET outperforms basic asynchronous processes in a use-case in which performance is positively correlated with evaluation time, and performs comparably (and often better) than basic asynchronous processes in several use-cases where performance is negatively correlated with evaluation time. That said, in the cases where performance and evaluation time are negatively correlated the variance of outcomes for SWEET is notably high.

关键词： asynchronous algorithms evolutionary algorithms optimization parallel computation

来源：评论

学校读者我要写书评

暂无评论

Performances analysis of parallel asynchronous and synchronous algorithms on cloud architecture for PDE

引用

ADVANCES IN ENGINEERING SOFTWARE 2023年 186卷

作者： Rahhali, M. A. Garcia, T. Spiteri, P. ENGIE La Defense Ile De France France CY Paris Univ CY Tech Campus Pau 2 Blvd Lucien Favre CS 77563 F-64075 Pau France Univ Toulouse IRIT INPT ENSEEIHT 2 Rue Charles CamichelBP 7122 F-31071 Toulouse 7 France

In order to solve discretized stationary linear or non linear diffusion and convection-diffusion problems, the present paper deals with a comparison between the performances of parallel iterative synchronous or asynchronous algorithms implemented on a Cloud computing architecture with several sizes of the algebraic systems to be solved. The parallel algorithms are briefly and intuitively presented and analysis of their behavior is recalled. In particular, criteria of convergence are recalled and estimates of convergence speed are given. Implementation on cloud architectures are briefly described. Finally, for each kind of problem, the performance of the target algorithms on cloud architectures are analyzed.

关键词： High performance computing Parallel computing Cloud computing asynchronous iterations asynchronous algorithms Iterative methods

来源：评论

学校读者我要写书评

暂无评论

SIMPLE, EFFICIENT, asynchronous PARALLEL algorithms FOR MAXIMIZATION

引用

ACM TRANSACTIONS ON PROGRAMMING LANGUAGES AND SYSTEMS 1988年第2期10卷 313-337页

作者： GREENBERG, AG LUBACHEVSKY, BD ODLYZKO, AM AT&T Bell Laboratories Murray Hill NJ AT&T Bell Laboratories Murray Hill NJ AT&T Bell Laboratories Murray Hill NJ

The problem of computing the maximum of n inputs on an asynchronous parallel computer is considered. In general, the inputs may arrive staggered in time, the number of processors available to the maximization algorithm may vary during its execution, and the number of inputs, n, may be initially unknown. Two simple, efficient algorithms to compute the maximum are presented. Each algorithm may be invoked asynchronously, as new inputs and processors arrive. Performance measures that account for the response times of the invocations are introduced, and the algorithms are analyzed under these measures.

关键词： asynchronous algorithms maximization response times

来源：评论

学校读者我要写书评

暂无评论

An asynchronous proximal bundle method

引用

MATHEMATICAL PROGRAMMING 2025年第1-2期209卷 825-857页

作者： Fischer, Frank Johannes Gutenberg Univ Mainz Inst Comp Sci Mainz Germany

We develop a fully asynchronous proximal bundle method for solving non-smooth, convex optimization problems. The algorithm can be used as a drop-in replacement for classic bundle methods, i.e., the function must be given by a first-order oracle for computing function values and subgradients. The algorithm allows for an arbitrary number of master problem processes computing new candidate points and oracle processes evaluating functions at those candidate points. These processes share information by communication with a single supervisor process that resembles the main loop of a classic bundle method. All processes run in parallel and no explicit synchronization step is required. Instead, the asynchronous and possibly outdated results of the oracle computations can be seen as an inexact function oracle. Hence, we show the convergence of our method under weak assumptions very similar to inexact and incremental bundle methods. In particular, we show how the algorithm learns important structural properties of the functions to control the inaccuracy induced by the asynchronicity automatically such that overall convergence can be guaranteed.

关键词： Proximal bundle methods Parallel programming Convex optimization asynchronous algorithms

来源：评论

学校读者我要写书评

暂无评论

Parallel cloud solution of large algebraic multivalued systems

引用

APPLIED NUMERICAL MATHEMATICS 2025年 208卷 366-389页

作者： Rahhali, M. A. Garcia, T. Spiteri, P. Univ Toulouse IRIT INPT ENSEEIHT 2 Rue Charles CamichelBP 7122 F-31071 Toulouse 7 France CY Paris Univ CY Tech Campus Pau2 Blvd Lucien FavreCS 77563 F-64075 Pau France

The present paper deals with the resolution on cloud architecture of synchronous and asynchronous iterative parallel algorithms of stationary or evolution variational inequations formulated by a multivalued model. The performances of synchronous and asynchronous iterative parallel methods are compared with previous ones obtained on cluster or when grid architecture is used. Thanks to the properties of the algebraic systems resulting from problem discretization we are able to analyze the behavior of the iterative algorithm in particular the convergence and the speed of convergence. The implementation of the studied methods on cloud architecture is described. Then we present various applications in particular the solidification of steel in continuous casting, the cavity pressure calculation described by a problem subject to unilateral constraints and finally a financial problem modeled by American option pricing.

关键词： High performance computing Parallel computing Cloud computing asynchronous iterations asynchronous algorithms Iterative methods Multivalued problems

来源：评论

学校读者我要写书评

暂无评论

Primal-Dual Fixed Point algorithms Based on Adapted Metric for Distributed Optimization With Bounded Communication Delays

引用

IEEE TRANSACTIONS ON AUTOMATIC CONTROL 2025年第4期70卷 2212-2227页

作者： Su, Yanxu Wang, Qingling Sun, Changyin Anhui Univ Sch Artificial Intelligence Hefei 230601 Peoples R China Minist Educ Engn Res Ctr Autonomous Unmanned Syst Technol Hefei 230601 Peoples R China Southeast Univ Sch Automat Nanjing 210096 Peoples R China

This article concentrates on distributed optimization over networks with communication delays. Each subsystem in the network performs its local updates by using the information received from its neighbors, be it possibly outdated. The communication delays with respect to different neighbors are assumed to be arbitrary but bounded. The objective function consists of a twice differentiable coupling term and an aggregated private term. The private function of each subsystem is the sum of two possibly nonsmooth terms, one of which is composed of a linear mapping. We propose a primal-dual fixed point algorithm framework based on the adapted metric for two scenarios where the coupling among subsystems is only enacted by the global objective function and enforced both by the global objective function and the linear mapping. The adapted metric method utilizes an adequate quadratic approximation of the global objective function as the updating step-size to exploit the second-order information. Under some mild assumptions, the convergence of the proposed algorithms is rigorously analyzed based on the quasi-Fej & eacute;r monotonicity. The numerical simulation verifies the correctness and effectiveness of the proposed algorithms.

关键词： Delays Optimization Couplings Measurement Convergence Linear programming Vectors Synchronization Symmetric matrices Signal processing algorithms Adapted metric methods asynchronous algorithms distributed optimization primal-dual fixed point (PDFP) algorithms

来源：评论

学校读者我要写书评

暂无评论

A Parallel Dynamic asynchronous Framework for Uncertainty Quantification by Hierarchical Monte Carlo algorithms

引用

JOURNAL OF SCIENTIFIC COMPUTING 2021年第1期89卷 1-25页

作者： Tosi, Riccardo Amela, Ramon Badia, Rosa M. Rossi, Riccardo Int Ctr Numer Methods Engn Barcelona Spain Barcelona Supercomp Ctr Barcelona Spain Univ Politecn Cataluna Barcelona Spain

The necessity of dealing with uncertainties is growing in many different fields of science and engineering. Due to the constant development of computational capabilities, current solvers must satisfy both statistical accuracy and computational efficiency. The aim of this work is to introduce an asynchronous framework for Monte Carlo and Multilevel Monte Carlo methods to achieve such a result. The proposed approach presents the same reliability of state of the art techniques, and aims at improving the computational efficiency by adding a new level of parallelism with respect to existing algorithms: between batches, where each batch owns its hierarchy and is independent from the others. Two different numerical problems are considered and solved in a supercomputer to show the behavior of the proposed approach.

关键词： Monte Carlo Multilevel Monte Carlo asynchronous algorithms Distributed Computing High Performance Computing

来源：评论

学校读者我要写书评

暂无评论

asynchronous analysis of parallel dynamic programming algorithms

引用

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS 1996年第4期7卷 425-438页

作者： Lewandowski, G Condon, A Bach, E UNIV WISCONSIN MADISON WI 53706 USA

We examine a very simple asynchronous model of parallel computation that assumes the time to compute a task is random, following some probability distribution. The goal of this model is to capture the effects of unpredictable delays on processors, due to communication delays or cache misses, for example. Using techniques from queueing theory and occupancy problems, we use this model to analyze two parallel dynamic programming algorithms. We show that this model is simple to analyze and correctly predicts which algorithm will perform better in practice. The algorithms we consider are a pipeline algorithm, where each processor i computes in order the entries of rows i, i + p, and so on, where p is the number of processors;and a diagonal algorithm, where entries along each diagonal extending from the left to the top of the table are computed in turn. It is likely that the techniques used here can be useful in the analysis of other algorithms that use barriers or pipelining techniques.

关键词： parallel dynamic programming asynchronous algorithms parallel models of computation queueing theory analysis of algorithms

来源：评论

学校读者我要写书评

暂无评论

Fourier based design of asynchronous phase detection algorithms

Fourier based design of asynchronous phase detection algorit...

引用

Conference on Optical Measurement Systems for Industrial Inspection V

作者： Gomez-Pedrero, Jose A. Crespo, D. Quiroga, Juan A. Univ Complutense Madrid Fac Ciencias Fis Dept Opt Ciudad UnivS-N E-28040 Madrid Spain Indizen Opt Technol Madrid 28020 Spain

ISBN: (纸本)9780819467584

Phase detection is an important issue when dealing with optical metrology techniques for which the magnitude to be measured is encoded through the phase of a given fringe pattern. asynchronous phase detection techniques are employed when the rate of phase change (frequency) it is not known. These techniques always present a variable frequency response, in other words, their ability to recover properly the phase depends strongly on the local frequency. In many experiments, it is possible to have a rough knowledge about the range of frequencies involved. Therefore, it constitutes a great advantage to have a procedure to design an asynchronous demodulation method which is suited to a particular frequency response for a given experiment. In this way, we get a better behaviour against noise which leads to more accurate and reliable phase extraction. In this work we present a technique to design asynchronous demodulation algorithms with a desired frequency response using a Fourier-based technique. The method allows the design of algorithms with a limited algebraic error in the recovered phase which have better properties than standard asynchronous phase detection techniques as it is shown in numeric and real experiments.

关键词： phase detection asynchronous algorithms frequency response Fourier techniques

来源：评论

学校读者我要写书评

暂无评论

asynchronous level bundle methods

引用

MATHEMATICAL PROGRAMMING 2020年第1-2期184卷 319-348页

作者： Iutzeler, Franck Malick, Jerome de Oliveira, Welington UGA Lab J Kuntzmann Grenoble France CNRS Lab J Kuntzmann Grenoble France PSL Res Univ MINES ParisTech CMA Ctr Math Appliquees Sophia Antipolis France

In this paper, we consider nonsmooth convex optimization problems with additive structure featuring independent oracles (black-boxes) working in parallel. Existing methods for solving these distributed problems in a general form are synchronous, in the sense that they wait for the responses of all the oracles before performing a new iteration. In this paper, we propose level bundle methods handling asynchronous oracles. These methods require original upper-bounds (using upper-models or scarce coordinations) to deal with asynchronicity. We prove their convergence using variational-analysis techniques and illustrate their practical performance on a Lagrangian decomposition problem.

关键词： Nonsmooth optimization Level bundle methods Distributed computing asynchronous algorithms

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：