Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userHPC compact quasi-Newton algorithm for interface problems
39
0
0.0
(
0
)
Added by
Alfonso Santiago Ferrandez
Publication date
2020
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
In this work we present a robust interface coupling algorithm called Compact Interface quasi-Newton (CIQN). It is designed for computationally intensive applications using an MPI multi-code partitioned scheme. The algorithm allows to reuse information from previous time steps, feature that has been previously proposed to accelerate convergence. Through algebraic manipulation, an efficient usage of the computational resources is achieved by: avoiding construction of dense matrices and reduce every multiplication to a matrix-vector product and reusing the computationally expensive loops. This leads to a compact version of the original quasi-Newton algorithm. Altogether with an efficient communication, in this paper we show an efficient scalability up to 4800 cores. Three examples with qualitatively different dynamics are shown to prove that the algorithm can efficiently deal with added mass instability and two-field coupled problems. We also show how reusing histories and filtering does not necessarily makes a more robust scheme and, finally, we prove the necessity of this HPC version of the algorithm. The novelty of this article lies in the HPC focused implementation of the algorithm, detailing how to fuse and combine the composing blocks to obtain an scalable MPI implementation. Such an implementation is mandatory in large scale cases, for which the contact surface cannot be stored in a single computational node, or the number of contact nodes is not negligible compared with the size of the domain. c{opyright} <2020> Elsevier. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/
rate research
Read More
In this paper, we discuss the problem of minimizing the sum of two convex functions: a smooth function plus a non-smooth function. Further, the smooth part can be expressed by the average of a large number of smooth component functions, and the non-smooth part is equipped with a simple proximal mapping. We propose a proximal stochastic second-order method, which is efficient and scalable. It incorporates the Hessian in the smooth part of the function and exploits multistage scheme to reduce the variance of the stochastic gradient. We prove that our method can achieve linear rate of convergence.
Recurrent Neural Networks (RNNs) are powerful models that achieve exceptional performance on several pattern recognition problems. However, the training of RNNs is a computationally difficult task owing to the well-known vanishing/exploding gradient problem. Algorithms proposed for training RNNs either exploit no (or limited) curvature information and have cheap per-iteration complexity, or attempt to gain significant curvature information at the cost of increased per-iteration cost. The former set includes diagonally-scaled first-order methods such as ADAGRAD and ADAM, while the latter consists of second-order algorithms like Hessian-Free Newton and K-FAC. In this paper, we present adaQN, a stochastic quasi-Newton algorithm for training RNNs. Our approach retains a low per-iteration cost while allowing for non-diagonal scaling through a stochastic L-BFGS updating scheme. The method uses a novel L-BFGS scaling initialization scheme and is judicious in storing and retaining L-BFGS curvature pairs. We present numerical experiments on two language modeling tasks and show that adaQN is competitive with popular RNN training algorithms.
This paper describes an extension of the BFGS and L-BFGS methods for the minimization of a nonlinear function subject to errors. This work is motivated by applications that contain computational noise, employ low-precision arithmetic, or are subject to statistical noise. The classical BFGS and L-BFGS methods can fail in such circumstances because the updating procedure can be corrupted and the line search can behave erratically. The proposed method addresses these difficulties and ensures that the BFGS update is stable by employing a lengthening procedure that spaces out the points at which gradient differences are collected. A new line search, designed to tolerate errors, guarantees that the Armijo-Wolfe conditions are satisfied under most reasonable conditions, and works in conjunction with the lengthening procedure. The proposed methods are shown to enjoy convergence guarantees for strongly convex functions. Detailed implementations of the methods are presented, together with encouraging numerical results.
In this paper, we propose some new proximal quasi-Newton methods with line search or without line search for a special class of nonsmooth multiobjective optimization problems, where each objective function is the sum of a twice continuously differentiable strongly convex function and a proper convex but not necessarily differentiable function. In these new proximal quasi-Newton methods, we approximate the Hessian matrices by using the well known BFGS, self-scaling BFGS, and the Huang BFGS method. We show that each accumulation point of the sequence generated by these new algorithms is a Pareto stationary point of the multiobjective optimization problem. In addition, we give their applications in robust multiobjective optimization, and we show that the subproblems of proximal quasi-Newton algorithms can be regarded as quadratic programming problems. Numerical experiments are carried out to verify the effectiveness of the proposed method.
Large, complex, multi-scale, multi-physics simulation codes, running on high performance com-puting (HPC) platforms, have become essential to advancing science and engineering. These codes simulate multi-scale, multi-physics phenomena with unprecedented fidelity on petascale platforms, and are used by large communities. Continued ability of these codes to run on future platforms is as crucial to their communities as continued improvements in instruments and facilities are to experimental scientists. However, the ability of code developers to do these things faces a serious challenge with the paradigm shift underway in platform architecture. The complexity and uncertainty of the future platforms makes it essential to approach this challenge cooperatively as a community. We need to develop common abstractions, frameworks, programming models and software development methodologies that can be applied across a broad range of complex simulation codes, and common software infrastructure to support them. In this position paper we express and discuss our belief that such an infrastructure is critical to the deployment of existing and new large, multi-scale, multi-physics codes on future HPC platforms.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
