Preparing, publishing, and using the Digital Typhoon Dataset for machine learning. The dataset is the longest typhoon satellite image dataset available. We developed ways to interact with the dataset for ML, and performed benchmarks in analysis, reanalysis, and forecasting.
Building tools with ML for analyzing neutrino detector images from a Liquid Argon Time Projection Chamber. Using a VQVAE to compress large image files, and performing instance segmentation on the LArTPC images to extract neutrino trails.
Developing a simulation code from the ground up for Radio Frequency Quadrupoles (RFQs). Designed to support existing simulation infrastructure, accurately simulate space-charge (interaction between charged particles), and parallelization. Faster and more scientifically accurate than existing alternatives!