NERSC AI
AI is transforming science across all kinds of domains and application areas within the DOE science portfolio. The computational needs of scientists developing AI applications are also growing. NERSC supports this evolving workload through
- Deployment of advanced hardware and software systems for AI
- Applications of AI for science through deep engagements with domain scientists
- Empowerment of the scientific community and workforce development through outreach events
NERSC has driven the emergence of modern AI and deep learning for science in recent years. Some highlights include:
- Built the first deep learning application to run on over 10k nodes with scientific tasks across climate and LHC physics (presented at SC17). This was followed by the NERSC-led first exascale deep learning application that won the 2018 Gordon Bell Prize.
- Deploying Perlmutter. In 2021, NVIDIA described this as the worlds fastest AI supercomputer at the time. This was quickly made available for Open Science. Early applications included the first deep learning model to achieve the skill of numerical weather prediction and novel particle physics publications
- Several first-of-a-kind deep learning applications were led by NERSC. Including the first generative deep learning for science (CosmoGAN and CaloGAN) in 2017, and the first self-supervised deep learning (for cosmology applications in 2020 and 2021). Further recent application publications with deep NERSC involvement are listed below.
- Running tutorials and schools to empower the community. For example the deep learning at scale tutorial at the SC conference has been led by NERSC since 2018 (SC23 material available here). Overall NERSC tutorials and schools have had 1000s of total participants, bringing AI expertise to the wider science and HPC world.
We are hiring!
The NERSC AI team currently has an opening for a Machine Learning Engineer to support the NERSC ML/DL software stack, deploy new cutting-edge tools & frameworks for scalable ML/DL workflows, and work with scientists to apply ML/DL techniques to their research.
The NERSC AI ecosystem
NERSC provides powerful computing systems for science, including our current flagship Perlmutter supercomputer, which is well designed for AI with over 7,000 NVIDIA A100 GPUs.
NERSC also provides a rich software ecosystem for AI, including prebuilt software environments, containers, and fully-customizable user environments.
Other relevant offerings for AI users include a JupyterHub service, the Spin platform for user-defined services, and the Superfacility API for interacting with NERSC systems in integrated and automated ways.
Related activities
Berkeley Lab AI for Science Summit
The Berkeley Lab AI for Science Summit, Oct. 24 and 25 at the Berkeley Residence Inn (Thu) and LBNL (Friday), will bring AI researchers, industry experts, scientists, and national lab staff together to explore how AI can drive scientific discoveries. The workshop will feature advancements in AI, focusing on scientific and engineering challenges and ethical and safety issues related to AI. More information and register at blass.ai Read More »
Fair Universe NeurIPS Competition Launched
The Fair Universe collaboration—composed of Berkeley Lab (NERSC and Physics Divisions), the University of Washington, Universite Paris-Saclay, and ChaLearn— is building an open, large-compute-scale AI ecosystem for benchmarks and challenges. The collaboration just launched their latest competition focussed on discovering and minimizing systematic uncertainties for experiments at the Large Hadron Collider. This is an accepted NeurIPS competition and winners will be invited to a special session at the conference in December. Learn more and join the challenge: fair-universe.lbl.gov
Read More »
NERSC Generative AI for Science Program
NERSC invited proposals for projects that will leverage Perlmutter to push the state of the art in Generative AI (GenAI) and deep learning for science, as well as produce novel science outcomes. The call was focused on teams with expertise in the use of deep learning for science, a thorough understanding of the scientific domain, and demonstrated proofs of concept. Read More »
StormCast: kilometer-scale weather with generative deep learning
NERSC, in a research collaboration led by NVIDIA, has recently published a preprint describing a new generative AI weather model called StormCast. StormCast is the first AI-driven weather model capable of emulating the atmosphere at mesoscale (~kilometer-scale) resolution, enabling the ability to represent convective systems and the evolution of extreme weather phenomena associated with them. These include events like thunderstorms, tornadoes, derechos, and extreme precipitation, which are of… Read More »
MLPerf HPC
With the explosion of interest in and exploration of AI in the DOE science communities, HPC centers are preparing for a shift toward new AI-enhanced computational workflows. It is imperative that the scientific HPC community be ready to support this emerging workload with representative and robust benchmarks that allow characterization of the computational workload and drive innovation in system and software design. MLPerf benchmarks from the MLCommons organization are the industry standard… Read More »
Huge Ensembles for Extreme Weather Prediction
Huge ensembles for extreme weather (HENS), a large collaboration between NERSC, LBNL EESA, NVIDIA, and Indiana University done as part of NERSC’s NESAP program has demonstrated an innovative new approach to ensemble weather forecasting that leverages deep learning models to produce much larger ensembles than what has been possible previously. Traditionally, numerical models require massive computational resources to simulate large ensembles of forecasts—key for accurately capturing the… Read More »
NERSC Postdoc Publishes OmniLearn: a Foundation Model for Jets in High Energy Physics
NERSC postdoctoral researcher Vinicius Mikuni, part of the NESAP program, has published an innovative paper titled OmniLearn: A Method to Simultaneously Facilitate All Jet Physics Tasks, currently under peer review. The research introduces OmniLearn, a machine learning-based approach that advances multiple high energy physics challenges by leveraging a single, versatile model. Jet physics, central to high-energy particle collisions, presents complex challenges due to its high-dimensional… Read More »
Reconstructing Lyα Fields from Low-resolution Hydrodynamical Simulations with Deep Learning
A collaboration between NERSC and LBNL cosmologists has introduced an innovative deep learning approach that overcomes limitations of traditional hydrodynamical simulations of the lyman-alpha forest. By combining physics-driven simulations with generative neural networks, the approach is capable of producing outputs comparable to simulations with eight times higher resolution. This leap in resolution is achieved without the immense computational resources typically required, enabling accurate… Read More »
NERSC AI Publications
NERSC is deeply involved in several projects to push the state-of-the-art in deep learning for science. Our engineers, postdocs, and interns collaborate with scientists from a wide range of domains including high energy physics, climate and weather modeling, chemistry/materials, and biosciences. We work with multi-disciplinary teams from LBNL, external institutions, and industry. Read More »