Statistics-informed neural networks for multidimensional stochastic processes
Developing recurrent neural-network surrogates trained through ensemble statistics. Incorporating marginal and cross-variable statistics extends SINN to multidimensional stochastic processes without requiring trajectory-wise correspondence.
Graduate Student Researcher, UC Merced · Fall 2022–present. First-author manuscript under review, 2026.
SINN architecture comparison
Conducting a comparison of LSTM, GRU, Transformer, and Mamba architectures for Statistics-Informed Neural Networks. Manuscript in preparation.
Presented at SIAM NCC26, UC Davis, October 2026; accepted for SIAM MDS26, Salt Lake City, November 2026. Poster (NCC26 / MDS26, PDF)
Stochastic dynamical systems with a slowly varying parameter
Extending SINN to stochastic systems with slowly varying parameters using a branch–trunk recurrent architecture to reproduce parameter-dependent stochastic trajectories.
Ongoing work · Invited CSE27 talk.
Hybrid simulation at Lawrence Berkeley National Laboratory
Embedded SINN as a neural surrogate for the fast stochastic component of a two-way coupled particle–continuum hybrid simulation. Worked on integrating generated stochastic trajectories into the multiscale workflow for computationally efficient simulation.
Research Intern, Applied Mathematics and Computational Research Division · Summer 2025. Mentor: Dr. Andy Nonaka.
Collaborative research projects
Contributed to DOE's Machine-Learning-Based Surrogate Modeling for Stochastic Multiscale Simulation Methodology (DE-SC0024240) and NSF's LEAPS-MPS: Stochastic Particle-Continuum Hybrid Simulation Method for Model Heterogeneous Catalysts under Reaction Conditions (CHE-2213368).