In my internship at the Naval Nuclear Laboratory, I conducted research on Bayesian optimization for low and high-dimensional non-convex functions. Using scalable HPC workflows, I designed and executed approximately 100 numerical experiments comparing the efficiency, stability and reliability of an array combinations of ML models, acquisition functions, and sampling techniques. The resulting data can be used to guide the creation of optimal models and inputs for the optimization of engineering analyses. Additionally, I explored theoretical and algorithmic improvements that may speed up these processes. These findings were documented via visualizations and writeups and archived the data in a repository. My code was contributed to a production project following modern software development practices at the end of my internship.
Unfortunately, much of the work I did is classified due to the sensitive nature of nuclear labs. As such, while I would be happy to talk about what I learned and accomplished, some details will have to be withheld.
However, here you can find some of the work I am approved to share here: Kernel Inversions. It regards researching algorithmic improvements to some of the matrix inversion steps in the Bayesian Optimization code used.