深度学习编程框架  

Programming frameworks for deep learning algorithms

作　　者：王秉睿 兰慧盈 陈云霁[2] WANG Bingrui;LAN Huiying;CHEN Yunji(University of Science and Technology of China,Hefei 230026,China;Institute of Computing Technology Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学技术大学,安徽合肥230026 [2]中国科学院计算技术研究所,北京100190 [3]中国科学院大学,北京100049

出　　处：《大数据》2018年第4期56-63,共8页Big Data Research

基　　金：国家重点研发计划基金资助项目(No.2017YFB1003101);国家自然科学基金资助项目(No.61472396;No.61432016;No.61473275;No.61522211;No.61532016;No.61521092;No.61502446;No.61672491;No.61602441;No.61602446;No.61732002;No.61702478);北京市科技计划基金资助项目(No.Z151100000915072);中国科学院STS计划;国家重点基础研究发展计划("973"计划)基金资助项目(No.2015CB358800)~~

摘　　要：近年来,深度学习算法日益流行,在各种领域的应用都取得了出色的效果,受到工业界和学术界的广泛关注。越来越多的研究者开始利用深度学习算法解决实际问题(如图像分类、图像识别、语音识别、自然语言处理等)。人们提出了各种各样的深度学习编程框架,便于研究者们开发新的深度学习算法。这些深度学习库的设计原则、抽象层次各有不同。对常见的深度学习编程框架进行了分类介绍,针对深度学习编程框架设计中的关键问题进行了分析,并且对未来深度学习编程框架的发展方向做了展望,为以后深度学习编程框架的设计提供了设计思路和方向。In recent years, deep learning algorithms became increasingly pervasive. It has drew extensive attentions from both researchers and industries, as it achieves very promising results on many applications of various fields. More and more researchers began to use deep learning algorithms to solve practical problems（e.g., image classification, image recognition, speech recognition, and natural language processing）. Many deep learning frameworks and libraries were proposed so that researchers can develop new deep learning algorithms in a more convenient fashion. These frameworks and libraries were different in many aspects（e.g., design principles and abstraction）. Firstly, several pervasive deep learning frameworks were introduced, and then the critical issue of designing such frameworks was analyzed. At last, the future challenges of designing deep learning frameworks were discussed. The study provides ideas and directions for future design.

关 键 词：机器学习 编程框架 大数据 

分 类 号：F830[经济管理—金融学]