High-performance Support Vector Machines for Data-Intensive Applications
EMSL Project ID
29393
Abstract
Support vector machines (SVM) are a software technology with broad applicability for creating binary classifiers using training data. In much the same way that SCALAPACK efficiently enables many application areas that rely heavily on linear algebra, we are applying SVM's to perform large-scale binary classification in problem spaces such as remote protein homology and peptide identification. Through the proposed work we will profile a collection of publically available and home-grown SVM implementaitons focusing on efficiency of hardware utilization, time to solution, and statistical quality of solutions.
Project Details
Project type
Limited Scope
Start Date
2008-01-31
End Date
2008-03-04
Status
Closed
Released Data Link
Team
Principal Investigator
Team Members
Related Publications
Oehmen, CS and BM Webb-Robertson (2008) Evaluating the Computational Requirements of using SVM software to train Data-Intensive Problems. Nova Science Publishers, New York, accepted.