2010 Poster Sessions : CAMPAIGN - Clustering Algorithms in Modular, Parallel, and Accelerated Implementation for GPU Nodes

Student Name : Kai Kohlhoff
Advisor : Russ Altman
Research Areas: Artificial Intelligence
Data clustering algorithms are useful in many fields of computer science and, increasingly, the life sciences. For instance, data analysis in modern day biology and bioinformatics often requires the extraction of meaningful health-related patterns from sources such as gene expression data and protein structures. This can require substantial amounts of processing time putting a limit on the amount and quality of information that can be derived in reasonable time. To allow the development of more sophisticated analysis protocols, we present ‘CAMPAIGN’, a library of GPU-accelerated clustering algorithms for large-scale data sets. Equipped with an initial set of tools and GPU-ports of well-established algorithms, including k-means, k-centers, hierarchical clustering, and self-organizing map, CAMPAIGN is intended to form the basis for devising new parallel clustering techniques specifically tailored to GPU architectures. Initial work focuses on Nvidia’s ‘CUDA’ parallel computing engine. The library is kept modular to allow easy modification and extensibility, for example by future modules using OpenCL. Our first benchmarks show one to two orders of magnitude performance improvements of CAMPAIGN over CPU reference implementations and popular software packages such as MatLab.

Kai is a postdoctoral researcher in Stanford's Bioengineering department jointly advised by Profs. Russ B. Altman and Vijay Pande. Previous to his appointment at Stanford, Kai completed both a PhD in Structural Bioinformatics and an MPhil in Computational Biology at the University of Cambridge, UK. His first degree was a triple-major BSc in Computer Science, Bioinformatics, and Biology from Jacobs University in Germany. Among his various interests are graphics processor programming and molecular dynamics simulations of proteins.