大数据下的深度学习研究

摘要

给出了大数据和机器学习的子领域——深度学习的概念,阐述了深度学习对获取大数据中的有价值信息的重要作用.描述了大数据下利用图像处理单元(GPU)进行并行运算的深度学习框架,对其中的大规模卷积神经网络(CNN)、大规模深度置信网络(DBN)和大规模递归神经网络(RNN)进行了重点论述.分析了大数据的容量、多样性、速率特征,介绍了大规模数据、多样性数据、高速率数据下的深度学习方法.展望了大数据背景下深度学习的发展前景,指出在不远的将来,大数据与深度学习融合的技术将会在计算机视觉、机器智能等多个领域获得突破性进展.%The concepts of big data and deep learning (a subfield of machine learning) were given, and the importance of deep learning in acquiring valuable information from big data was interpreted.The deep learning framework for concurrent computation using graphics processing unit was described, and its big convolutional neural network (CNN), big deep belief network (DBN) and big recurrent neural network (RNN) were emphatically introduced.The features of big data in volume, variety and velocity were analyzed, and the methods for deep learning under large scale data, variable data and high rate data were introduced.The future development of the research on deep learning under big data was forecasted, and the possibility that the technology of fusing big data and deep learning will make an important breakthrough in the fields such as computer vision and machine intelligence was pointed out.