深度学习研究与进展  

Research and Advances on Deep Learning

作　　者：孙志远[1,2] 鲁成祥[1,3] 史忠植[1] 马刚[1,2] 

机构地区：[1]中国科学院计算技术研究所智能信息处理重点实验室,北京100190 [2]中国科学院大学,北京100049 [3]曲阜师范大学信息科学与工程学院,日照276826

出　　处：《计算机科学》2016年第2期1-8,共8页Computer Science

基　　金：国家"九七三"重点基础研究计划(2013CB329502);国家自然科学基金(61035003)资助

摘　　要：深度学习是机器学习领域一个新兴的研究方向,它通过模仿人脑结构,实现对复杂输入数据的高效处理,智能地学习不同的知识,而且能够有效地解决多类复杂的智能问题。近年来,随着深度学习高效学习算法的出现,机器学习界掀起了研究深度学习理论及应用的热潮。实践表明,深度学习是一种高效的特征提取方法,它能够提取数据中更加抽象的特征,实现对数据更本质的刻画,同时深层模型具有更强的建模和推广能力。鉴于深度学习的优点及其广泛应用,对深度学习进行了较为系统的介绍,详细阐述了其产生背景、理论依据、典型的深度学习模型、具有代表性的快速学习算法、最新进展及实践应用,最后探讨了深度学习未来值得研究的方向。Deep learning（DL）is a recently-developed field belonging to machine learning.It tries to mimic the human brain,which is capable of processing the complex input data fast,learning different knowledge intellectually,and solving different kinds of complicated human intelligence tasks well.Recently,with the advent of a fast learning algorithm for DL,the machine learning community set off a surge to study the theory and applications of DL since it has many advantages.Practice shows that deep learning is a kind of high efficient feature extraction method,which can detect more abstract characteristics and realize the essence of the data,and the model constructed by DL tends to have stronger generalization ability.Due to the advantages and wide applications of deep learning,this paper attempted to provide a started guide for novice.It presented a detailed instruction of the background and the theoretical principle of deep learning,its emblematic models,its representative learning algorithm,the latest progress and applications.Finally,some research directions of deep learning that are deserved to be further studied were discussed.

关 键 词：深度学习 机器学习 深层神经网络 图像识别 语音识别 自然语言处理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]