深度学习(Deep Learning)综述及算法简介（Overview of Deep Learning and algorithm introduction）.doc

深度学习(Deep Learning)综述及算法简介（Overview of Deep Learning and algorithm introduction） This article is translated from deeplearning website, which mainly summarizes some papers and algorithms. Introduction: Neural Network and Support Vector Machines (SVM) are representative methods of statistical learning. Neural networks and support vector machines can be considered to be derived from Perceptron. The perceptron is a linear classification model developed by Rosenblatt in 1958. Perceptrons are valid for linear classification, but the classification problems in reality are usually nonlinear. Neural networks and support vector machines (including nuclear methods) are nonlinear classification models. In 1986, Rummelhart and McClelland invented the neural network learning algorithm Back Propagation. Subsequently, Vapnik et al. introduced support vector machines in 1992. The neural network is a nonlinear model of multiple layers (usually three layers). The support vector machine USES nuclear techniques to transform nonlinear problems into linear problems. Neural networks and support vector machines have been in a competitive relationship. Scholkopf was a great disciple of Vapnik, who supported the research of vector machines and nuclear methods. According to Scholkopf, Vapnik invented the support vector machine to kill Neural Network. Support vector machines are indeed very effective, and support vector machines have gained the upper hand over time. In recent years, Hinton, a master of neural network, has proposed the Deep Learning algorithm for neural networks (2006), which greatly improves the neural network capability, which can be compared with support vector machines. Deep Learning assumes that the neural network is multilayered, and first, Boltzman Machine (unsupervised Learning) is used to learn the structure of the network, and then the weight of the network is learned through Back Propagation (supervised Learning). As for the name of Deep Learning, Hinton joked: I want