自动化学报  2017, Vol. 43 Issue (11): 1886-1897   PDF    
深度学习认知计算综述
陈伟宏1,2, 安吉尧1,2, 李仁发1,2, 李万里1,2     
1. 湖南大学信息科学与工程学院 长沙 410082;
2. 嵌入式与网络计算湖南省重点实验室 长沙 410082
摘要: 随着大数据和智能时代的到来,机器学习的研究重心已开始从感知领域转移到认知计算(Cognitive computing,CC)领域,如何提升对大规模数据的认知能力已成为智能科学与技术的一大研究热点,最近的深度学习有望开启大数据认知计算领域的研究新热潮.本文总结了近年来大数据环境下基于深度学习的认知计算研究进展,分别从深度学习数据表示、认知模型、深度学习并行计算及其应用等方面进行了前沿概况、比较和分析,对面向大数据的深度学习认知计算的挑战和发展趋势进行了总结、思考与展望.
关键词: 深度学习     认知计算     张量数据表示     并行计算     大数据    
Review on Deep-learning-based Cognitive Computing
CHEN Wei-Hong1,2, AN Ji-Yao1,2, LI Ren-Fa1,2, LI Wan-Li1,2     
1. College of Computer Science and Electronic Engineering, Hunan University, Changsha 410082;
2. Key Laboratory for Embedded and Network Computing of Hunan Province, Changsha 410082
Manuscript received : October 10, 2016, accepted: June 22, 2017.
Foundation Item: Supported by National Natural Science Foundation of China (61672217, 61370097)
Author brief: CHEN Wei-Hong Ph. D. candidate at the College of Computer Science and Electronic Engineering, Hunan University, and professor at Hunan City University. She received her master degree from Hunan University in 2006. Her research interest covers cyber physical systems, distributed computing, and machine learning;
LI Ren-Fa Professor at Hunan University. He received his Ph. D. degree from Huazhong University of Science and Technology. His research interest covers embedded system, cyber-physical systems, artificial intelligence, and machine vision;
LI Wan-Li Ph. D. candidate at the College of Computer Science and Electronic Engineering, Hunan University. He received his bachelor degree from Hunan University in 2014. His research interest covers machine learning, computer vision, intelligent transportation systems, and driver behavior analysis
Corresponding author. AN Ji-Yao Professor at Hunan University. He received his Ph. D. degree from Hunan University in 2012. His research interest covers cyber-physical systems, parallel and distributed computing, and computing intelligence. Corresponding author of this paper
Recommended by Associate Editor ZHANG Hua-Guang
Abstract: With the advent of the era of big data and artificial intelligence, the research focus of machine learning has shifted from perception domain to cognitive computing (CC) domain. How to improve the cognitive ability through big data is becoming a research hotspot of intelligence science and technology, in which recent deep learning has been expected to spark a new wave of research on cognitive computing. This paper summarizes the research progress of cognitive computing based on deep learning in recent years. And, comparison and analysis of recent progress in deep learning and cognitive computing are presented from three aspects, that is, deep learning data representation, cognitive models, parallel computing and its applications in the big data environment. Finally, some challenges and development trends of cognitive computing based on deep learning for big data are investigated to for cast the future research.
Key words: Deep learning (DL)     cognitive computing (CC)     tensor data representation     parallel computing     big data    

认知计算(Cognitive computing, CC)源于模拟人脑的计算机系统的人工智能, 是通过人与自然环境的交互及不断学习, 帮助决策者从不同类型的海量数据中揭示非凡的洞察, 以实现不同程度的感知、记忆、学习和其他认知活动[1].随着大数据时代的到来, 丰富的数据和知识为认知计算迎来了新的机遇.与此同时, 数据的规模、种类、速度和复杂度都远远超过了人脑的认知能力, 如何有效完成对大数据的认知, 给传统认知计算也带来了巨大挑战[2].

认知计算是对新一代智能系统特点的概括.从功能层面上讲, 认知系统具备人类的某些认知能力, 能够出色完成对数据的发现、理解、推理、决策等特定认知任务[3].认知计算是解决理解和学习的问题, 学习能力是认知系统的关键, 特别是在当前大数据时代, 可供学习的数据和知识越来越丰富.近年来, 得益于计算机硬件性能的提升和云计算技术的发展, 深度学习(Deep learning, DL)作为一种新的机器学习方法, 已成为大数据时代认知计算的研究热点之一[4].深度学习通过构建基于表示的多层机器学习模型, 训练海量数据, 学习有用特征, 以达到提升识别、分类或预测的准确性[5].深度学习可以超越概念学习, 学习到更加复杂的知识, 是深层神经网络学习算法的重大突破[6-7].深度学习认知计算是基于深度学习方法挖掘数据中的价值, 以得到更准确、更深层次的知识, 提升对数据的认知能力[8-9].基于深度学习的围棋程序AlphaGo已达到职业棋手水平[10].

目前, 一个深度学习认知计算过程主要包含三个方面:深度学习数据表示、深度学习认知模型、深度学习并行计算, 本文综述为三个模块, 如图 1所示.认知计算的目标是从输入数据中通过模型学习和高性能计算实现分类和理解等认知活动.大数据环境下, 数据具有大量、多样性、异构性等特征, 如何有效表示数据将直接影响认知模型的建立以及认知计算的效果, 数据表示是深度学习认知计算的基础.与传统的浅层网络相比, 深层网络模型具有更强大的学习能力, 好的深度学习认知模型将能得到好的认知效果, 模型是关键.随着大数据集的空前增长, 数据复杂多变, 以及深度模型的复杂, 认知算法处理的是NP难问题, 单机计算能力远不能满足深度学习训练的需要, 高性能并行计算成为实现深度学习认知计算的保障.

图 1 深度学习认知计算示意图 Figure 1 Diagram of deep learning cognitive computing

由于深度学习认知算法和计算能力的突破, 深度学习掀起了认知计算领域的一次革命.目前, 国内外深度学习的研究机构诸如多伦多大学、蒙特利尔大学、斯坦福大学、纽约大学、微软研究院、Google、IBM研究院、百度公司等, 已将深度学习成功应用于计算机视觉[11]、手写体识别[12]、图像和语音识别[13-15]、音频处理[16-