Research Area: Machine Learning
In recent years, the breakthrough of deep learning in the field of artificial intelligence algorithms has triggered an academic upsurge which attracted more and more researchers. As a multi-layer perceptron, the key to its success lies in the local link and weight-sharing method. On the one hand, it reduces the quantity of weights and makes the network easier to optimize. On the other hand, it reduces the risk of over-fitting. A weight-sharing network’s structure of the convolutional neural network makes it more similar to a biological neural network, which reduces the complexity of the network model and quantity of weights. In the processing of image problems, especially recognizing displacement, scaling, and other forms of distortion invariant applications, it has better robustness and operation efficiency. First of all, this paper reviews the development history of convolutional neural network. Secondly, it introduces the basic structure of convolutional neural network, and elaborates its differences from ordinary artificial neural networks in terms of operating principles. It also analyzes the details of convolutional neural network’s structural framework which includes convolutional layers, subsampling layers, and fully connected layers. Finally, the advantages of convolutional neural network in image processing, speech analysis, and other fields are given at last.
Keywords:
Author(s) Name: Xv Zhang, Chenxi Xv, Ming Shen, Xin He, Wei Du
Journal name:
Conferrence name: International Conference on Network, Communication, Computer Engineering
Publisher name: Atlantis press
DOI: 10.2991/ncce-18.2018.16
Volume Information: volume 147
