Brandon Gusto | Principal Research Scientist

About

I am a Principal Research Scientist and Department of Defense SMART Scholar. My expertise spans modeling and simulation, optimization algorithms, software development, machine learning, and signal processing.

My research goes beyond developing new algorithms. I architect and deliver production-quality scientific software built for high performance computing. These software often depend on MPI, performance portable GPU libraries, and popular scientific computing libraries.

Expertise

Scientific Computing

Development of scalable, high-performance software for numerical simulations and data analysis pipelines.

Multifidelity Bayesian Optimization

Novel optimization schemes combining multi-objective optimization with multifidelity surrogate models for efficient design space exploration.

Computational Fluid Dynamics

High-fidelity CFD simulations for engineering applications.

Acoustic Signal Processing

Algorithm development for acoustic analysis, beamforming, source localization, and noise characterization in complex environments.

Skills

Object-Oriented C++ Fortran Python Matlab MPI CUDA OCCA MAGMA SLATE PETSc HDF5 PyTorch Adaptive Mesh Refinement Large-Eddy Simulation Reynolds-Averaged Navier-Stokes Bayesian Optimization Gaussian Processes Software Engineering High Performance Computing Dynamic Mode Decomposition

Projects

Selected research projects and software contributions demonstrating applied expertise.

Multi-Fidelity Bayesian Optimization

Optimization + ML

Extended random embedding Bayesian optimization to leverage multi-fidelity data sources for high-dimensional problems. Introduced a multi-armed bandit approach for intelligent embedding selection using Thompson sampling.

Matlab Python

DDT in Type Ia Supernovae

CFD + ML

Large-scale DNS campaign (27,791 simulations) exploring deflagration-to-detonation transition in thermonuclear supernovae. Developed neural network classifiers achieving 97%+ accuracy for predicting detonation formation from hotspot configurations.

Fortran Python TensorFlow MPI

Hybrid Adaptive Multiresolution

CFD + HPC

Novel mesh adaptation method combining block-structured AMR with multiresolution analysis for reactive flow simulations. Achieves significant speedups through solver-adaptive techniques that replace expensive calculations with interpolation in smooth regions.

Fortran MPI AMR FLASH