Biography
Andrew joined Berkeley Lab in 2013 as a postdoctoral researcher in the Applied Numerical Algorithms Group. He is now a staff member in the Center for Computational Sciences and Engineering, where he designs and implements HPC algorithms for solving multiscale, multiphysics problems using structured adaptive meshes and particles. He is a core contributor to the AMReX adaptive mesh library, and also contributes to a number of AMReX-based simulation codes in subjects ranging from computational plasma physics to epidemiology. He was a member of the 2022 Gordon-Bell prize-winning team for kinetic plasma simulations with the WarpX Particle-in-Cell code, and received a Director's Award for Exceptional Scientific Achievement in 2023.
SRP Project Title
Python-driven workflows with AMReX
HPSF Project
AMReX
Topical Areas
Applied Computer Science; Applied Mathematics; High Performance Computing; Software Engineering
Abstract
AMReX is a publicly available software framework designed for building massively parallel block- structured adaptive mesh refinement (AMR) applications. Simulation codes based on AMReX model a wide range phenomena from fields ranging from astrophysics and cosmology to plasma physics, earth systems modeling, multi-phase flow, epidemiology, cell biology, and more. While AMReX is written in C++, for this internship we are envisioning several projects that involve improving and expanding the Python interfaces to AMReX with the goal of supporting new AI/ML use cases, including: 1. Integrating simulations with ML-based Bayesian optimization workflows 2. Training fast surrogate models and incorporating those into simulations 3. Automatic differentiation of coupled C++ simulation and Python analysis code through tools like Enzyme. 4. Uncertainty quantification using frameworks like PyTUQ. The above workflows will be demonstrated on and evaluated with real-world AMReX-based application codes.
Desired Skills
Experience with Python is desired, other useful skills include experience with C++, AI/ML, and high-performance computing.
Lightning Talk Title
AMReX: Adaptive Mesh Refinement for Exascale
Keywords
Adaptive Mesh Refinement; Math libraries; Scientific Computing; C++; Python; AI/ML; automatic differentiation; uncertainty quantification