Effective and Efficient Data Analysis Techniques for Emerging Data Intensive Applications

从手持设备到传感器、监控到微阵列再到Web的广泛可用性(例如，社交媒体、社交网络、物联网、点击流等)技术的最新发展。允许以电子方式捕获、收集和维护大量结构化和非结构化数据，从而形成大数据存储库。示例数据源包括与通过Web访问的信息检索相关的日志和事务、诸如点击流和零售购买的电子商务、诸如天气和股市状态的正在进行的事件、诸如社交媒体和社交网络的社交交互、诸如交通控制、欺诈检测和国土安全等监控系统等。这些数据被认为是知识发现的宝贵资源，有助于制定有效的决策。然而，只有在能够有效地处理和分析这些海量数据的情况下，才能实现这一点，而用传统方法来处理和分析这些数据越来越成问题。有时，在数据实例数量与拥有数百个样本和数千个基因等大量特征相比较少的情况下，数据可能会出现偏差；这可以通过数据丰富来解决。此外，真实数据通常遭受各种其他相互关联的问题，这些问题需要处理降维、缺失值和噪声、在有限计算能力下的可扩展性(即，使用常见的计算机)、数据建模和集成以揭示特定语义(例如，捕获和监控社交媒体数据中的行为和趋势以允许更好的推荐)等。这些问题不能孤立地处理，因此需要能够在数据中存在时处理这些问题的任何组合的集成框架。 这个研究项目的主要目标是通过开发一个集成了高效和健壮的网络建模、数据挖掘和机器学习技术的框架来导致有效和信息丰富的知识发现。学生将参与方法论的各个部分，从开发必要的理论和算法到实施和测试它们。

Recent developments in the technology from handheld devices to sensors to surveillance to microarrays to the wide availability of the Web (e.g., social media, social networking, internet of things, click stream, etc.) allow for electronically capturing, collecting and maintaining huge volumes of structured and unstructured data leading to big data repositories. Example data sources include logs and transactions related to information retrieval via Web access, to e-commerce such as click streams and retail purchases, on-going events such as weather and stock market status, social interactions such as social media and social networking, monitoring systems such as traffic control, fraud detection, and homeland security, and so on. Such data is recognized as a valuable resource for knowledge discovery leading to effective decision making. Yet this can happen only if these large volumes of data can be processed and analyzed effectively, which is increasingly problematic to do by conventional means. Sometimes the data could be skewed where the number of data instances is small compared to the huge number of features like having hundreds of samples and thousands of genes; this could be fixed by data enrichment. In addition, real data generally suffers from various other interrelated problems that require dealing with dimensionality reduction, missing values and noise, scalability under limited computing power (i.e., using commonly available computers), data modeling and integration to uncover specific semantics (e.g., capturing and monitoring behavior and trend in social media data to allow for better recommendations), etc. These problems cannot be treated in isolation and hence there is a need for an integrated framework capable of handling any combination of these problems when present in the data. The main objective of this research program is leading to effective and informative knowledge discovery by developing a framework which integrates efficient and robust network modeling, data mining and machine learning techniques. Students will be involved in various parts of the methodology from developing the necessary theory and algorithms to implementing and testing them.