Application Of Neural Network And Deep Learning In Quantitative Investment

Google Alphago in weiqi game overwhelmingly defeated famous nine play-ers Li Shishi news has aroused great concern, artificial intelligence and depth of learning also started for more people to understand. Deep learning is pro-posed on the basis of artificial neural network, it is contains a hidden layer of the neural network, compared to a simple neural network it has more powerful ability to learn, to the data to make a higher level of abstraction. Therefore, the practical application effect is better. But the depth of the neural network training is always a difficult problem. Before 2006, the depth of the neural net-work is rarely in the machine learning field is brought up, because it is good training, and it is easy to over fitting leading to poor generalization ability. Until 2006, Professor Hinton put forward the concept of deep learning, and through the layer by layer training method to solve the problem of the depth of the neural network is difficult to train. After that, Hinton and others pro-posed to solve the neural network over fitting of the Dropout method, so that the depth of the network’s ability to improve the generalization. At present, deep learning has been widely applied to the field of image recognition, speech recognition and Natural Language Processing, and set off a boom of research.In this paper, the theory of neural network and deep learning is introduced and summarized, and the theoretical model is tried to apply to financial data to test its practical application effect. The full text is divided into four chap-ters:the first chapter briefly introduces neural network and depth of learning development and the organizational structure; the second chapter, the paper introduce a simple neural network and its some improvement methods; barri-ers to learning in the third chapter, we introduce depth and depth of learning models. Chapter four will be learning neural network and the depth method is applied to stock index futures data and comparative analysis of the two effects. From the most basic perceptron model began to introduce the neural network, after the introduction of sigmoid neurons and neural network, introduced the neural network and then introduces the training methods and the most com-monly used is the most basic error back propagation algorithm. Then it will analyze some of the problems in the neural network and introduce the corre-sponding improvement methods for specific problems. After the introduction of simple neural network, this paper will introduce a specific example to in-troduce the depth of neural network training difficulties, and then it will be introduced to make these difficulties have been successfully solved the depth of learning model. Among them will be mentioned restricted Boltzmann ma-chines, it is a deep belief network based and the will to it are described in detail. Also talked about another in image recognition in the common depth learning model convolutional neural network model. In this paper will be the core idea in detail. After the introduction of the theoretical part, the final at-tempt to simple neural network and depth of learning model is applied to the stock index futures closing price prediction, the input model will use a trading day and before the transaction price and transaction data,and the output is a trading day closing price of ups and downs, if rising output is 1 if the decline is 0, training model after we enter new data, and comparing the model output with real data, in order to determine the model application effect respectively, and of simple neural network and depth of learning model to compare with the effect.