Data Intensive Computing
Data Intensive Machine Learning for Real-Time Decision Analysis
Principal Investigators: Bobbie-Jo Webb-Robertson, Chris Oehmen
Challenge
Analyze large data samples accurately and quickly in real time
Approach
- Use existing software and high performance computing platforms
- Develop robust vectorization strategies for new domains
- Develop generalized fusion strategies for pair-wise SVM comparisons
Impact
Tackle problems in domain spaces requiring
- High sensitivity
- Real-time analysis at the sensor
- Analysis of large complex datasets
Collaborations
- Proteomics informatics group at EMSL, a DOE national user facility)
- National security clients
Accomplishments
- Evaluation: SVM training algorithms can perform in near real-time on large data (19K vectors)
- Demonstrated on data intensive bioinformatics applications
Relative speedup of training on about 19K vector dataset using parallel SVM on multiple processors