high performance computing on graphics processing units: hgpu.org

In this thesis, we present a framework for efficient analysis of large-scale network datasets. There are four important components in our framework: a) a high performance computing platform with Graphics Processing Units (GPUs) and efficient implementations of mining algorithms on top of the GPU platform. b) an efficient summarization method to compress the storage space of large-scale streaming and heterogeneous network data with textual content and network topology. c) a complex query engine that depends on the summarized input data and can help to discover new knowledge from network content and topology. d) a visual front-end to present mining results to users. In each of the above components, we propose new methods to either speed up mining tasks or reduce the data storage size in those tasks. We compare our methods with existing approaches on real datasets drawn from various domains.