PNNL

Recently, the U.S. Department of Energy’s Office of Advanced Scientific Computing Research awarded project funding for “Convergence of Deep Learning and Machine Learning for HPC Modeling and Simulation.” Abhinav Vishnu, research scientist with PNNL’s High Performance Computing group, will serve as the project’s principal investigator, overseeing the effort, which will focus on developing extreme-scale Deep Learning algorithms. The project team also includes Charles Siegel (Applied Statistics and Computational Modeling Group) and Jeff Daily, Shuaiwen Leon Song, Joseph Manzano, and Darren Kerbyson (all from PNNL HPC).

Top row (l-to-r): Vishnu (PI), Daily, and Siegel. Bottom row (l-to-r): Song, Manzano, and Kerbyson. Enlarge Image.

The project’s main objective is to develop algorithms that are capable of executing on novel compute, memory, and storage architectures. The methods uncovered during the course of the research will be applied to several domains of interest to DOE, including discovery of new particles and light sources.

“It is the convergence of big data and big compute, which is a reason for the success of Deep Learning in science,” Vishnu said. “A significant innovation for further advancing Deep Learning will come from high-performance computing, and this project will play a critical role in achieving that objective.” 

The research prototype for scalable Deep Learning is available as part of the Machine Learning Toolkit for Extreme Scale (MaTEx), available at: http://hpc.pnl.gov/matex/ and https://github.com/abhinavvishnu/matex.