Thanks for visiting my site!
I am a PhD candidate in Statistics at Oregon State University. Broadly speaking, my research interests lie in
- predictive modeling
- computational statistics
- machine learning
for complex (1-D) time series modeling, with applications to energy forecasting and claims prediction. My other interests include deep learning and its more flexible and explainable probabilistic variant models (e.g., deep Gaussian processes, deep Gaussian Markov Random Fields) for (2-D) space and (3-D) space-time modeling.
PhD
For my PhD dissertation, advised by Dr. Lisa Madsen and Dr. Charlotte Wickham (now at Posit, PBC), I focus on the Markov MTD (Mixture Transition Distribution) models and the Bayesian NNMP (Nearest-Neighbor Mixture Process) models for time series forecasting and spatial prediction, as developed by Zheng et al. and presented in [1], [2], and [3]. To achieve flexible and scalable modeling of time series and spatial data, both the MTD and NNMP models utilize probabilistic graphical model (DAG or Bayesian network) representations, finite mixture model structures, Bayesian hierarchical model formations, and MCMC algorithms.
My specific contributions include incorporating the copulas in the MTD model to capture more complex dependence structures; extending the MTD model to handle zero-inflated skewed data; validating the proposed models and assessing model performance through simulations and real-world data applications; developing a software package for the proposed models (using R, C++); and comparing the proposed models with other benchmark approaches, such as LSTM (Long Short-Term Memory) networks (using PyTorch).