University of New Mexico | Jan 2026 – Present | Albuquerque, NM
As a Research Assistant, I work on building machine learning performance models for heterogenous applications running on LLNL’s Tuolumne supercomputer.
What I’ve Done
- Developed machine learning models built with PyTorch and XGBoost to predict scaling behavior of AMG2023 and other GPU accelerated applications on Tuolumne.
- Designed and integrated methods for uncertainty quantification, active learning, and modeling performance variation. This improved prediction reliability and accuracy, and reduced data collection costs.
- Automated performance data collection for heterogeneous applications running on Tuolumne at Lawrence Livermore National Laboratory. Utilized the Flux scheduler, Spack, and DuckDB. Modified source code of applications to add Caliper instrumentation and capture process topology.
Technologies
Python, C/C++, PyTorch, XGBoost, Scikit-Learn, Polars, Caliper, Flux, Linux, DuckDB, Cray-MPICH (MPI), ROCm