Project-Based Research and Training in High-Performance Data Sciences, Data Analytics, and Machine Learning

Kwai Wong, Stanimire Tomov, and Jack Dongarra

Volume 11, Issue 1 (January 2020), pp. 36–44

PDF icon Download PDF

  author={Kwai Wong and Stanimire Tomov and Jack Dongarra},
  title={Project-Based Research and Training in High-Performance Data Sciences, Data Analytics, and Machine Learning},
  journal={The Journal of Computational Science Education},
Copied to clipboard!

This paper describes a hands-on project-based Research Experiences for Computational Science, Engineering, and Mathematics (RECSEM) program in high-performance data sciences, data analytics, and machine learning on emerging computer architectures. RECSEM is a Research Experiences for Undergraduates (REU) site program supported by the USA National Science Foundation. This site program at the University of Tennessee (UTK) directs a group of ten undergraduate students to explore, as well as contribute to the emergent interdisciplinary computational science models and state-of-the-art HPC techniques via a number of cohesive compute and data intensive applications in which numerical linear algebra is the fundamental building block. The RECSEM program complements the growing importance of computational sciences in many advanced degree programs and provides scientific understanding and discovery to undergraduates with an intellectual focus on research projects using HPC and aims to deliver a real-world research experience to the students by partnering with teams of scientists who are in the forefront of scientific computing research at the Innovative Computing Laboratory (ICL), and the Joint Institute for Computational Sciences (JICS) at UTK and Oak Ridge National Laboratory (ORNL). The program also receives collaborative support from universities in Hong Kong and Changsha, China. The program focuses on scientific domains in engineering applications, image processing, machine learning, and numerical parallel solvers on supercomputers and emergent accelerator platforms, particularly their implementation on GPUs. The programs also enjoy close affiliations with researchers at ORNL. Because of these diverse topics of research areas and backgrounds of this project, in this paper we discuss the experiences and resolutions in managing and coordinating the program, delivering cohesive tutorial materials, directing mentorship of individual projects, lessons learned, and improvement over the course of the program, particularly from the perspectives of the mentors.