CAREER: Memory-Constrained Predictive Data Mining

Temple University, Philadelphia PA

Investigators

Abstract

In the environment where new large-scale problems are emerging in various disciplines and pervasive computing applications are becoming more common, there is an urgent need for techniques that provide efficient and accurate knowledge discovery by limited-capacity computing devices. The objective of this project is to address this need by developing memory-constrained predictive data mining algorithms that operate when data size exceeds the available memory capacity. The approach is based on the integration of data mining and data compression techniques to optimally utilize the memory for data and model storage, learning, and ancillary operations. The methodology will be thoroughly evaluated on a range of real-life problems that includes learning from large sorted databases, biased data and nonstationary data. Various memory constraints will be considered pertaining to devices ranging from powerful workstations to handheld computers and cell phones to small, inexpensive sensors. This research will reveal the memory lower bounds for accurate learning from different types of data and by different types of learning algorithms. The educational component of the project seeks to integrate research into computer science instruction by designing exciting courses, exploring effective teaching techniques, introducing research to undergraduate and graduate students, and involving underrepresented student groups in research. Broader impacts of the project will be in extending the frontiers of computer and information science and in facilitating knowledge discovery in various scientific, engineering, and business disciplines. Teaching materials and research results, including developed software and databases, will be widely disseminated via Internet (http://www.ist.temple.edu/~vucetic/CAREER.htm) to promote learning and enhance scientific understanding.