SOLAR：服务导向的深度学习架构

182 浏览量更新于2024-08-27 收藏 618KB PDF 举报

"SOLAR: Services-Oriented Learning Architectures" 在当前的计算机科学领域，深度学习已经成为机器学习的一个重要分支，特别是在过去的几十年里。然而，随着数据的多样化和规模的扩大，构建灵活且高效深度学习神经网络的实现面临着重大挑战。为了提升性能并保持可扩展性，"SOLAR: Services-Oriented Learning Architectures"这篇研究论文提出了一种创新的解决方案。 SOLAR是一种服务导向型的深度学习架构，它利用了如GPU和FPGA等硬件加速器来优化深度学习的执行效率。这个架构的主要目标是提供一个统一的编程模型，使用户可以专注于算法设计，而无需关注底层硬件实现和调度策略。这样，程序员可以更专注于模型的开发，而复杂的加速器管理和任务调度则由系统自动处理。在运行时，SOLAR能够根据需求将服务执行在软件处理器或硬件加速器上，这极大地提高了计算资源的利用率和整体系统的响应速度。通过这样的方式，SOLAR能够有效地应对大规模数据集和复杂模型带来的计算压力，同时也支持动态调整以适应不断变化的工作负载。实验结果表明，SOLAR的架构能显著提高深度学习任务的性能，同时保持良好的可扩展性。这不仅减少了开发者的负担，也为深度学习在各种应用场景中的部署提供了更高效、更灵活的途径，比如图像识别、自然语言处理、自动驾驶等领域。通过这种方式，SOLAR为服务导向型的学习架构设定了新的标准，为未来深度学习系统的优化和设计提供了新的思路。此外，SOLAR可能还涉及到以下几个关键知识点： 1. **服务化编程模型**：这种模型使得开发者可以像调用服务一样使用深度学习模型，简化了开发流程，并降低了使用高级硬件加速器的门槛。 2. **硬件加速器集成**：SOLAR整合了GPU和FPGA等硬件加速技术，通过智能调度机制，能够在不同硬件平台上实现模型的无缝迁移和优化执行。 3. **性能优化与可扩展性**：通过服务化架构，SOLAR能够动态调整资源分配，确保在数据量增加或模型复杂度提升时仍能保持高性能。 4. **异构计算**：SOLAR支持在软件处理器和硬件加速器之间进行任务分配，利用异构计算的优势，提高了整体计算效率。 5. **实时性和灵活性**：由于其服务化的特性，SOLAR能够快速响应变化的需求，提供实时的深度学习服务。 "SOLAR: Services-Oriented Learning Architectures"这篇研究论文为深度学习的实施提供了一个创新的框架，它通过服务化和硬件优化，解决了大规模数据和复杂模型的处理难题，对深度学习领域的研究和实践具有重要的指导价值。

SOLAR: Services-oriented Learning Architectures

Deep Learning as a Service

Chao Wang

, Xi Li

, Qi Yu

, Aili Wang

, Patrick Hung

, and Xuehai Zhou

School of Computer Science, University of Science and Technology of China

Business and Information Technology, University of Ontario Institute of Technology

cswang,llxx,wangal,xhzhou@ustc.edu.cn, yuiq@mail.ustc.edu.cn, patrick.hung@uoit.ca

Abstract— Deep learning has been an emerging field of ma-

chine learning during past decades. However, the diversity and

large scale data sizes have posed significant challenge to con-

struct a flexible and high efficient implementations of deep

learning neural networks. In order to improve the perfor-

mance as well to maintain the scalability, in this paper we pre-

sent SOLAR, a services-oriented deep learning architecture

using various accelerators like GPU and FPGA based ap-

proaches. SOLAR provides a uniform programming model to

users so that the hardware implementation and the scheduling

is invisible to the programmers. At runtime, the services can be

executed either on the software processors or the hardware

accelerators. Experimental results on the real state-of-the-art

FPGA board demonstrate that the SOLAR is able to provide a

ubiquitous framework for diverse applications without increas-

ing the burden of the programmers. Moreover, the speedup of

the GPU and FPGA hardware accelerator in SOLAR can

achieve significant speedup comparing to the conventional

Intel i5 processors with great scalability.

Keywords- Services-oriented Architecture; Deep Learning;

neural network; accelerator.

I. INTRODUCTION

In the past few years, machine learning has become per-

vasive in various research fields and commercial applications,

and achieved satisfactory products. In particular, the emer-

gence of deep learning speeded up the development of ma-

chine learning and artificial intelligence. Consequently, deep

learning has become a research hotspot in research organiza-

tions and the companies [1]. In general, deep learning uses a

multi-layer neural network model to extract high-level fea-

tures into a combination of low-level abstractions to find the

distributed data features, in order to solve complex problems

in machine learning. Currently the most widely used neural

models of deep learning are Deep Neural Networks (DNNs)

and Convolution Neural Networks (CNNs) [2], which have

excellent capability in solving picture recognition, voice

recognition, bioinformatics, and other complex machine

learning tasks [3].

However, with the increasing accuracy requirements and

complexity for the practical applications, the size of the neu-

ral network is becoming explosively large scale, which

makes the data centers quite power consuming. Therefore, it

poses significant challenges to implement high performance

deep learning networks with low power cost, especially for

large-scale deep learning neural network models.

So far, the state-of-the-art means for accelerating deep

learning algorithms include Field-Programmable Gate Array

(FPGA), Application Specific Integrated Circuit (ASIC), and

Graphic Processing Unit (GPU) based approaches. Of these

approaches, GPU has been well recognized for its high per-

formance in massive computing capacity. Compared with

GPU acceleration, hardware accelerators like FPGA and

ASIC can achieve at least moderate performance with lower

power consumption. However, both FPGA and ASIC have

relatively limited computing resources, memory, and I/O

bandwidths, therefore it is difficult to develop complex and

massive deep neural networks using hardware accelerators.

Up to now, the problem of providing efficient middleware

support for different architectures has not been properly

solved.

Another challenge is diversity and programming in deep

learning applications. Due to the design complexity of the

deep learning algorithms, architectures and accelerators, it

requires significant programming effort to make satisfying

utilization of the accelerations in diverse application domains.

If the complex computation is solved by the programmer

manually, the quality of scheduling depends on the experi-

ences of the programmer, who has limited knowledge of the

hardware. In order to alleviate the burden of the high level

programmers, we can demonstrate the effectiveness of ser-

vices-oriented architecture (SOA) in deep learning research

paradigm. Traditionally, SOA provides effective measures

with better flexibility and extensibility at lower cost through

adopting re-usable modules in software engineering. To the

best of our knowledge, there have been no researches that

utilize SOA concepts to solve the diversity of the deep learn-

ing applications.

To tackle this problem, in this paper we present SOLAR,

which is a services-oriented deep learning architecture tar-

geting the state-of-the-art machine learning applications. Our

main contributions are the following:

1. We introduce SOA concepts into the state-of-the-art

architecture for deep learning applications. Services are pro-

vided by heterogeneous accelerators within a well-structured

interface to improve the flexibility and scalability.

2. SOLAR is based on a heterogeneous hybrid system

which includes the software processor, GPU accelerator, and

hardware FPGA accelerator to speed up the kernel computa-

tional parts of deep learning algorithms. In particular, we

utilize an efficient middleware support to bridge the gap be-

tween high level neural networks and hardware accelerators.

2016 IEEE International Conference on Web Services

DOI 10.1109/ICWS.2016.91

662

下载后可阅读完整内容，剩余3页未读，立即下载

weixin_38630612

粉丝: 5
资源: 891

SOLAR：服务导向的深度学习架构

You Only Cache Once: Decoder-Decoder Architectures for Language

pattern-oriented-software-architectures:coursera 上 POSA 课程的作业

Security for Web Services and Service-Oriented Architectures.pdf

Networkers2009：BRKBBA-2007 - Evolution of SP Edge Architectures

Networkers2009：BRKBBA-2015 - NGN Service Interconnect and SIP trunking architectures and scenarios

Switch／Router Architectures：Shared-Bus and Shared-Memory Based Systems 2018.pdf

DLNA Guidelines Part 1-1: Architectures and Protocols --2014

Android-Base-Google-Architectures:从Google Architectures，我正在做基础项目

TF-NAS:TF-NAS

p4fpga：P4-1416 Bluespec编译器

最新资源