Oak Ridge Leadership Computing Facility Enables Breakthrough Science
Published on Thursday 2 January 2020
By: Jack Wells, Director of Science, Oak Ridge National Laboratory National Center for Computational Sciences
The Oak Ridge Leadership Computing Facility was established at Oak Ridge National Laboratory over 25 years ago. We set out on a mission to accelerate scientific discovery and engineering progress by providing world-leading computational performance and advanced data infrastructure.
One key to our success in this mission has been our partnership with the OpenPOWER Foundation. Collaboration between industry leaders including IBM Power Systems, Nvidia, Mellanox and more enabled the creation of Summit, the world’s most powerful supercomputer since June 2018.
As director of science for the Oak Ridge Leadership Computing Facility, it’s been a joy to oversee the scientific outcomes of our user program, many of which are using groundbreaking artificial intelligence and deep learning technologies, and have incredible potential to improve the world as we know it. Don’t just take my word for it; learn more about four research projects that have each been conducted on Summit below.
Deep Learning Expands Study of Nuclear Waste Remediation
A team from Lawrence Berkeley National Laboratory, Pacific Northwest National Laboratory and NVIDIA has achieved exaflop performance on Summit with a deep learning application used to model subsurface flow in the study of nuclear waste remediation. This work demonstrates the promise of physics-informed generative adversarial networks (GANs) for analyzing complex, large-scale science problems.
Results from the study were presented at SC19. Learn more about the project here.
Artificial Intelligence Approach Points to Bright Future for Fusion Energy
A team of researchers led by Bill Tang of the Princeton Plasma Physics Laboratory and Princeton University tested their Fusion Recurrent Neural Network (FRNN) code on Titan and Summit. Using neural networks, FRNN identifies patterns in plasma behavior to quickly and accurately predict disruptions in fusion reactors.
According to Tang, “with powerful predictive capabilities, we can move from disruption prediction to control, which is the holy grail in fusion. It’s just like in medicine - the earlier you can diagnose a problem, the better chance you have of solving it.” Learn more about this project here.
AI for Plant Breeding in an Ever-changing Climate
Dan Jacobson, a research and development staff member in the Biosciences Division at Oak Ridge National Laboratory and his team is currently working on numerous projects that form an integrated roadmap for the future of AI in plant breeding and bioenergy. They recently developed a new genomic selection algorithm driven by explainable AI and expanded to a global scale the climate and environmental information that can be used in the Combinatorial Metrics, or CoMet, code.
You can find a Q&A with Jacobson on the project here.
In the Fight Against Cancer, ORNL and Stony Brook Cancer Center Enlist and Advanced Neural Network
Using the MENNDL code on Summit, an ORNL team has created a multi-objective neural network that can speed up cancer pathology research by using neural networks that can quickly and accurately analyze biopsy slide images on a scale that microscope-equipped pathologists could never completely tackle.
According to Joel Saltz, chair of the Department of Biomedical Informatics and associate director of the Stony Brook Cancer Center, “tumors are a little like stealth aircraft - they manage to actively confuse the patient’s immune system in order to not be recognized and killed.” Read more about the project here.
First MD Simulation Trajectories Transformed into Images Recognized by Deep Learning Technology
A team led by Harel Weinstein, D.Sc. took 3D visual representations of molecular dynamics data and transformed them into 2D picture-like representations and then trained a convolutional neural network to analyze and predict the class labels of the drugs or ligands that bind to two specific serotonin and dopamine receptors in humans with near-perfect accuracy.
The study builds a framework for the efficient computational analysis of MD big data collected for the purpose of understanding ligand-specific GPCR activity. Read more on the study here.
Which of these five projects do you believe contains the most significant potential to impact the world? I would love to hear your perspective in the comments section below!
**Note: Post updated to include final project, which was completed in February 2020.