Postdoc: Thermodynamic Computing for ML on GPUs

Lawrence Berkeley National Laboratory; San Francisco (CA); On-site; USD 99,000 - 111,000; Full time

Job summary

Lawrence Berkeley National Laboratory’s Molecular Foundry seeks a Postdoctoral Scholar in Thermodynamic Computing to scale physics-based computing using thermal noise as a resource. The postdoc will run large-scale simulations at NERSC and help design computers capable of performing state-of-the-art machine-learning tasks in collaboration with researchers at the Foundry and NERSC.

A thermodynamic computer uses fluctuating degrees of freedom to perform calculations; nonlinear, nonequilibrium

Qualifications

• PhD in physics, computer science, applied mathematics, or a related field.
• Experience developing numerical simulations of physical systems, or training machine-learning models.
• Proficiency in Python, C++, Julia, or Fortran.
• Ability to formulate research questions and pursue them independently.
• Strong communication skills both written and verbal with demonstrated ability to communicate results through publications or presentations.
• Experience with GPU computing and large-scale HPC resources.

Responsibilities

• Design and run large-scale simulations of nonlinear, nonequilibrium thermodynamic computers on GPUs at NERSC, using hybrid evolutionary and gradient-based training methods.
• Benchmark the accuracy, speed, and energy cost of thermodynamic computers against digital neural networks on standard machine-learning tasks.
• Determine how design choices, such as network connectivity and driving protocols, control inference accuracy and energy dissipation.
• Write and maintain documented, reproducible simulation code.
• Publish results in peer-reviewed journals, and present them at conferences and internal meetings.

Skills

Python

C++

Julia

Fortran

Machine learning

GPU computing

HPC resources

Education

PhD in physics, computer science, applied mathematics, or related field

Tools

PyTorch

JAX

Job description

Lawrence Berkeley National Laboratory’s Molecular Foundry seeks a Postdoctoral Scholar in Thermodynamic Computing to scale physics-based computing using thermal noise as a resource. The postdoc will run large-scale simulations at NERSC and help design computers capable of performing state-of-the-art machine-learning tasks in collaboration with researchers at the Foundry and NERSC.

A thermodynamic computer uses fluctuating degrees of freedom to perform calculations; nonlinear, nonequilibrium