大数据分析的神经网络方法

发布时间：2018-06-06 17:56

本文选题：大数据 + 神经网络　；参考：《工程科学与技术》2017年01期

【摘要】：大数据蕴含巨大的社会、经济、科学价值,已成为学术界与企业界关注的重点。其关键技术可划分为三大层次:数据平台、分析平台和展示平台,其中分析平台是大数据转化为价值的桥梁。一般来说,大数据拥有体量浩大(volume)、多源异构(variety)、生成快速(velocity)、价值稀疏(value)的"4V"特性,扩大了大数据的价值空间,同时也为大数据的分析技术带来巨大挑战。其中三大挑战比较显著,即多源异构大数据、大量非结构化数据存储、大数据价值稀疏且变化快。其三大核心科学问题为大数据的表达、存储和预测问题。由于传统的数据分析方法难以胜任,发展新的大数据分析方法势在必然。人脑是天然的大数据处理引擎。神经网络是一种模拟人脑大数据分析机制的计算方法,是目前大数据分析中最成功的方法。神经网络的研究主要包括:模拟大脑神经网络结构,构建神经网络结构模型;模拟大脑神经网络的记忆机制,发展学习算法。神经网络的研究历史历经波折。近年来,随着当代计算机计算能力的不断提升,基于神经网络的大数据分析方法取得了巨大成功,尤其是在各应用领域,如语音大数据分析、图像大数据分析、医学大数据分析等,引领了人工智能的发展。Alpha Go在人机围棋大战中获胜,引起了广泛关注。"大数据+神经网络"已成为驱动创新、推动社会发展和改变人类生产生活方式的一种重要力量。以大数据和神经网络为线索,回顾大数据的基本概念与关键技术,梳理神经网络研究的基本框架,可以发现它们之间默契切合、互相促进的关系。一方面,神经网络具有强大的特征提取与抽象能力,能够整合多源信息,处理异构数据,捕捉变化动态,是大数据实现价值转化的桥梁。另一方面,体量浩大的大数据为神经网络提供了充足的训练样本,使得训练越来越大规模的神经网络成为可能。尽管"大数据+神经网络"在众多应用领域已经取得了突破,但是,仍然存在需要解决的核心科学问题。面向神经网络的研究中,神经网络的结构尚需进一步研究,神经网络的大小依然缺少理论性的指导,神经网络的学习算法仍然存在一些内在的问题。围绕大数据分析的三大核心科学问题,需要研究如何保证在高维空间中稀疏表达仍可维持数据的一致性,如何实现"只存储知识而不存储原始数据",如何刻画数据的时空关联以实现大数据的预测。因此,仍然需要对该领域持续投入,加强应用研究和理论研究,尤其应进行跨领域的研究,即与人脑的大数据处理相呼应,结合认知科学、神经科学等相关学科的知识,以解决神经网络和大数据应用中的核心科学问题,推动基于神经网络方法的大数据分析研究。
[Abstract]:Big data contains enormous social, economic and scientific value, which has become the focus of academic and business circles. The key technology can be divided into three levels: data platform, analysis platform and display platform, in which the analysis platform is the bridge to transform big data into value. Generally speaking, big data has the "4V" feature of volume, multi-source heterogeneity, fast velocity.Value-sparse value), which expands the value space of big data, and also brings great challenges to the analysis technology of big data. The three major challenges are multi-source heterogeneous big data, large amount of unstructured data storage, and big data value is sparse and fast. Its three core scientific problems are the expression, storage and prediction of big data. Because the traditional data analysis method is not competent, it is necessary to develop a new big data analysis method. The human brain is a natural big data processing engine. Neural network is a method to simulate the mechanism of big data analysis in human brain. It is the most successful method in big data analysis at present. The research of neural network mainly includes: simulating the neural network structure, constructing the neural network model, simulating the memory mechanism of the neural network and developing the learning algorithm. The research history of neural network has gone through twists and turns. In recent years, with the continuous improvement of computer computing power, the neural network-based big data analysis method has achieved great success, especially in various application fields, such as voice big data analysis, image big data analysis. "Medical big data analysis has led the development of artificial intelligence. Alpha go has won the game of go, which has attracted widespread attention." Big data neural network has become an important force to drive innovation, promote social development and change the way of life and production of human beings. Based on the clue of big data and neural network, this paper reviews the basic concepts and key technologies of big data, combs the basic framework of neural network research, and finds out the relationship between them. On the one hand, neural network has powerful ability of feature extraction and abstraction, which can integrate multi-source information, process heterogeneous data and capture dynamic changes. It is a bridge for big data to realize value transformation. On the other hand, the large volume of big data provides sufficient training samples for neural networks, which makes it possible to train more and more large-scale neural networks. Although "big data neural network" has made a breakthrough in many application fields, there are still some core scientific problems to be solved. In the research of neural network, the structure of neural network needs further study, the size of neural network is still lack of theoretical guidance, and the learning algorithm of neural network still has some inherent problems. Focusing on the three core scientific problems of big data analysis, it is necessary to study how to ensure the consistency of data in sparse representation in high-dimensional space. How to realize "storing knowledge instead of storing raw data", and how to depict the spatio-temporal correlation of data to realize the prediction of big data. Therefore, it is still necessary to continue to invest in this field and to strengthen applied and theoretical research, especially to carry out cross-domain research, that is, to echo the big data processing of the human brain, and to combine the knowledge of cognitive science, neuroscience and other related disciplines. In order to solve the core scientific problems in the application of neural network and big data, the research of big data analysis based on neural network is promoted.
【作者单位】：四川大学计算机学院;
【基金】：国家自然科学基金重点项目资助(61432012;U1435213)
【分类号】：TP311.13;TP18

【相似文献】