Zhiyuan Jerald Wang (王志远)

I am currently pursuing a Ph.D. degree in the Department of Electricity and Computer Engineering at Texas A&M University supervised by Prof. Xiaoning Qian. I got my M.S. degree in software engineering at University of Electronic Science and Technology of China supervised by Prof. Fan Zhou.

My recent research interests include time series analysis, uncertainty, and deep generative models.

Email  /  Google Scholar  /  Github

[08/2023] I am employed as a research assistant at Texas A&M University.

[12/2022] I am awarded the title of "The Most Outstanding Students Award of UESTC" (Top 10 outstanding graduates).

[10/2022] I am awarded the title of "Outstanding Graduate of Sichuan Province".

[09/2022] I obtain China National Scholarship.

[12/2021] I am invited to serve as a PC member of WWW 2022 industrial track.

[08/2021] I am invited to serve as a reviewer of IEEE TII.

To better forecast multivariate time series forecasting, we present CGMF, which consists of a continuous graph module incorporating ODE to capture the long-range intra- and inter-relations. We also introduce a CDE-based fusion mechanism that exploits multi-scale representations to form continuous evolutional dynamics across different scales.

Learning domain-invariant representations is a major task of out-of-distribution generalization. To address this issue, we propose a novel model DyCVAE to learn dynamic causal factors. The results on synthetic and real datasets demonstrate the effectiveness of our proposed model for the task of generalization in time series domain.

Motivated by the success of disentangled variational autoencoder in computer vision and classical time series decomposition, we propose LaST that infers a couple of representations that depict seasonal and trend components of time series. Extensive experiments demonstrates its superiority on the time series forecasting task.

We propose a novel framework named GraphGeo, which first provides a complete processing methodology for street-level IP geolocation with the application of graph neural networks. It incorporates IP hosts knowledge and kinds of neighborhood relationships into the graph to infer spatial topology for high-quality geolocation prediction.

we propose a novel method named Probabilistic Electricity Forecasting (PrEF) by proposing a non-linear neural state space model (SSM) and incorporating copula-augmented mechanism into that, which can learn uncertainty-dependencies knowledge and understand interactions between various factors from large-scale electricity time series data.

Less is known about the scenario of an IP address, e.g., dedicated enterprise network or home broadband. In this work, we initiate the first attempt to address a large-scale IP scenario identification problem.

We present HydroFlow, a novel deep generative model for predicting the electricity generation demand of large-scale hydropower stations. It uses a latent stochastic RNN to capture the dependencies in the multivariate time series while considering the uncertainty of variables related to natural and social factors.