Recent Advances in Convolutional Neural Networks

卷积神经网络的最新进展

Abstract—In thelast few years, deep learning has led to very good performance on a variety ofproblems, such as visual recognition, speech recognition and natural languageprocessing. Among different types of deep neural networks, convolutional neuralnetworks have been most extensively studied. Due to the lack of training dataand computing power in early days, it is hard to train a large high-capacityconvolutional neural network without overfitting. After the rapid growth in theamount of the annotated data and the recent improvements in the strengths ofgraphics processor units (GPUs), the research on convolutional neural networkshas been emerged swiftly and achieved state-ofthe-art results on various tasks.In this paper, we provide a broad survey of the recent advances inconvolutional neural networks. Besides, we also introduce some applications ofconvolutional neural networks in computer vision.

近年来，深入学习在很多问题上表现出非常好的性能，例如视觉识别，语言识别以及自然语言处理。在众多类型的深度神经网络中，卷积神经网络又是被研究得最广泛的一个，由于早期缺少训练数据以及计算能力，很难训练一个大型的高能量的卷积神经网络而没有过拟合。随着标注数据的迅速增长以及图形处理器的迅速改善，卷积神经网络在很多任务上的研究很快出现，而且达到了最好的效果。本文对最新进展的进行了丰富的综述。同时我们也介绍了一些卷积神经网络在计算机视觉方面的应用。

这是卷积神经网络及其计算机视觉上面的应用的一个综述。Jiatao Guo 以及其他几位同学都新加坡南阳理工的团队的。这个综述非常专业，值得仔细学习与体会。