EFFICIENT MACHINE LEARNING METHODS FOR BIG DATA ANALYTICS

用于大数据分析的高效机器学习方法

• 批准号: RGPIN-2016-04855
• 负责人: LameirasKoerich, Alessandro
• 金额: $ 1.6万
• 依托单位: École de technologie supérieure
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 财政年份: 2016
• 资助国家: 加拿大
• 起止时间: 2016-01-01 至 2017-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=615339
• 关键词: EFFICIENT; MACHINE; LEARNING; METHODS; BIG

Currently, big data plays a key role in science and in industry. Big data concern large volume of complex and growing datasets coming from multiple sources. There is a growing demand for approaches that are able to explore large volumes of data and extract useful information or knowledge for future actions. This large amount of data currently available provides an unprecedented opportunity to extract information and to answer several questions that were previously considered intangible. However, several challenges must be met to unveil the great potential of big data analytics. Recent studies in machine learning that claim to be on big data actually used stationary data of low dimensionality where the decisions are taken on binary concepts. Big data is commonly unstructured, the data comes from diversified sources and the volume of data grows continuously. Besides the volume and the complexity of data, two other characteristics must be taken into account: data variety and velocity. The term volume is the size of dataset, velocity indicates the speed of data in and out, and variety describes the range of data types and sources. Therefore, there is a big gap between what has been delivered by current big data analytics approaches and the promises. There are several authors that claim to deal with big data analytics but they are limited to “built models on current stored data and use such models to predict on new data”. While they are able to handle large volumes of unstructured data, they have trouble to deal with heterogeneous and dynamically changing data. If the characteristics of data change, the learned model will fail to relate the observed data to a correct concept. The main goal of our research program is to fill such a gap by proposing and developing efficient machine learning methods for big data analytics that cope not only with the large volume of data, but also with velocity and variety inherent of big data. To achieve such a goal, our research program is structured along two main objectives: (i) development of learning methods adapted to unstructured and heterogeneous data on stationary data [volume and variety]; (ii) development of adaptive learning methods for non-stationary data [velocity]. This research program is fundamental to pave the way to the development of big data analytics applications that go beyond the data volume but also deal with the variety and the velocity of big data. The students enrolled in this research program will become specialists in data science and will be highly qualified to disseminate the advances in knowledge for the scientific community and Canadian industries and businesses.

目前，大数据在科学和工业中发挥着关键作用。大数据涉及来自多个来源的大量复杂且不断增长的数据集。人们越来越需要能够探索大量数据并为未来行动提取有用信息或知识的方法。目前可获得的大量数据为提取信息和回答以前被认为是无形的几个问题提供了前所未有的机会。然而，要揭示大数据分析的巨大潜力，必须应对几个挑战。 最近的机器学习研究声称是关于大数据的，实际上使用了低维度的静态数据，其中的决策是基于二进制概念的。大数据通常是非结构化的，数据来自不同的来源，数据量不断增长。除了数据的数量和复杂性之外，还必须考虑另外两个特征：数据种类和速度。术语卷是数据集的大小，速度表示数据进出的速度，多样性描述了数据类型和来源的范围。 因此，当前大数据分析方法所提供的内容与承诺之间存在很大差距。有几位作者声称处理大数据分析，但他们仅限于“在当前存储的数据上构建模型，并使用这些模型来预测新数据”。虽然他们能够处理大量的非结构化数据，但他们在处理异构和动态变化的数据时遇到了麻烦。如果数据的特征发生变化，学习的模型将无法将观察到的数据与正确的概念联系起来。 我们研究计划的主要目标是通过提出和开发用于大数据分析的有效机器学习方法来填补这一空白，这些方法不仅科普大量数据，还可以处理大数据固有的速度和多样性。为了实现这一目标，我们的研究计划是结构沿着两个主要目标：（i）学习方法的发展，适应非结构化和异构数据的固定数据[量和品种];（ii）自适应学习方法的发展，为非固定数据[速度]。 该研究计划是为大数据分析应用程序的开发铺平道路的基础，这些应用程序不仅涉及数据量，而且还涉及大数据的多样性和速度。参加该研究计划的学生将成为数据科学专家，并将有资格为科学界和加拿大工业和企业传播知识的进步。