I'm presenting our work, "Neural Diffusion Intensity Models for Point Process Data", at UAI 2026 in Amsterdam, the Netherlands.
Xinlong Du
I'm a PhD student in Industrial Engineering at Purdue University, working in the Stochastic System Lab led by Prof. Harsha Honnappa. Previously, I studied physics at Reed College and statistics at Purdue University.
I'm mostly interested in solving inference problems in stochastic systems using modern machine learning techniques, but I also enjoy thinking about problems in game theory, reinforcement learning, and differential privacy.
Currently, I'm working on temporal point process modeling and in-context reinforcement learning.
News
Projects
NDIM
Neural Diffusion Intensity Models for Temporal Point Process Data. This is joint work with Prof. Harsha Honnappa and Prof. Rao Vinayak. By identifying a common structure in the posterior dynamics of a diffusive intensity, we proposed a variational framework for training a prior model and an amortized posterior correction. This replaces the expensive MCMC simulations at inference time, while still retaining the posterior prediction accuracy.
SimBaRepro
Simulation-based, finite-sample inference for privatized data. I worked with Prof. Jordan Awan and Zhanyu Wang to turn their research paper into an R package published on CRAN.
