My research investigates the dynamics of relativistic particles within planetary radiation belts. As these environments are characterized by complex, multi-scale plasma processes, my work focuses on integrating traditional kinetic theory with emerging data science methodologies to improve both physical understanding and predictive capability.
Key methodological areas of interest include:
Physics-Informed Machine Learning: Developing Physics-Informed Neural Networks (PINNs) that embed governing plasma equations (such as the Fokker-Planck equation) directly into the learning process. This approach aims to ensure that model reconstructions remain physically consistent even when satellite data is sparse.
Predictive Space Weather Modeling: Utilizing recurrent architectures (LSTMs/GRUs) and Transformer models to ingest solar wind time-series data for the probabilistic forecasting of energetic electron enhancements.
Physics Discovery from High-Dimensional Data: Applying data mining and symbolic regression to large-scale satellite datasets (e.g., Van Allen Probes, THEMIS) to identify hidden relationships between wave-particle interactions and global particle transport.
Hybrid Simulation & Inference: Combining high-performance 3D particle tracing with Bayesian inference to better quantify physical parameters, such as radial diffusion coefficients, from observational signatures.
Through these efforts, we seek to advance geospace research by leveraging the University of Michigan’s data science and AI ecosystem to build more robust, interpretable models of the space environment.