Data summarization on probabilistic databases

点击翻译按钮获取中文摘要

Data summarization techniques have been used in many database management systems to provide quick approximate answers to queries when exact answers are not required or when early feedbacks are helpful. For example, the query optimizers in database management systems employ data summarization techniques to quickly estimate the sizes of intermediate results and the costs of different query execution plans. Many on-line analytical processing, data mining, and data visualization applications also rely on data summarization techniques to compute fast approximate answers to queries. Currently, data summarization techniques are generally applied to databases in which all data values are precise and accurate. However, in many databases, the data values are often associated with uncertainty and follow a probabilistic distribution. The uncertainty is possibly originated from inherent measurement inaccuracies, sampling errors, or network latencies. Such databases are referred to as the probabilistic databases in the literature. Probabilistic databases have been recently studied in a number of domains, including database management, sensor networks, consumer behavior modeling, and bioinformatics. The primary objective of this project is to develop new data summarization techniques on probabilistic databases. These techniques will be evaluated through experiments on synthetic and real data sets. Results from this project will significantly contribute toward probabilistic database management.