Grant: Discovery Grants Program - Individual

This proposal covers distributed data management issues in two increasingly important domains: XML query processing and data stream systems.- Distributed XML query processing: XML has emerged as the de-facto standard for publishing and exchanging data of many different types. While extensive research has been conducted on efficiently querying XML data in a wide range of settings, these have primarily focused on data that reside on a single machine (i.e., centralized). This has two potential problems. On the one hand the solutions may not be scalable to very large XML databases, while on the other hand, these approaches do not address the necessity to query and retrieve existing XML data that may already be distributed. In this research we will study distributed XML query processing in two main environments: (1) when there is some control over where the XML data can be placed (i.e., tightly controlled distribution), and (2) when data are already distributed (i.e., the data integration scenario). - Mining data streams: A data stream is a real-time, continuous, ordered (implicitly by arrival time or explicitly by timestamp) sequence of items. The particular problem that will be studied is the mining of data streams with distribution changes (i.e., ""dynamic data streams""). Traditional data mining assumes that data are entirely stored on disk, and almost all traditional mining algorithms require scanning the data multiple times. Neither of these assumptions hold in the case of streaming data. ""Distribution change"" refers to changes over time in the underlying statistical distribution of the data. For dynamic data streams, mining results generated in the past may not be relevant as time progresses. However, since data can be looked at only once, historical data are no longer available, and, thus, it is even hard to just determine the change point of the underlying distribution, let alone updating mining results to fit the new data. We will perform a systematic study of the problem focusing on effective (and statistically meaningful) detection of distribution changes and online mining algorithms to cope with these changes.