DIRECTION 3 · HPC
High Performance Computing
Making reconstruction and statistical analysis fast — CPU/GPU performance, parallelization, and search-based tuning.
Physics analyses are bounded by compute as much as by statistics. The lab profiles the hot path, vectorizes and parallelizes it, and moves the heavy fits onto the GPU — so reconstruction and likelihood scans run within the experiment’s computing budget instead of dictating it.
The work spans GPU statistical frameworks (the GooStats / GooFit-based fitter), CPU/GPU reconstruction speedups for the TAO detector, and search- and learning-based tuning — a Monte-Carlo-Tree-Search optimizer for configuration spaces, and a study of when and why the Muon optimizer helps.
Other research directions