Predictive modeling of spatial-temporal data: A graph-centric approach
Sun, Jiarui
This item is only available for download by members of the University of Illinois community. Students, faculty, and staff at the U of I may log in with your NetID and password to view the item. If you are trying to access an Illinois-restricted dissertation or thesis, you can request a copy through your library's Inter-Library Loan office or purchase a copy directly from ProQuest.
Permalink
https://hdl.handle.net/2142/129495
Description
Title
Predictive modeling of spatial-temporal data: A graph-centric approach
Author(s)
Sun, Jiarui
Issue Date
2025-02-25
Director of Research (if dissertation) or Advisor (if thesis)
Chowdhary, Girish
Doctoral Committee Chair(s)
Chowdhary, Girish
Committee Member(s)
Schwing, Alexander Gerhard
Driggs-Campbell, Katherine
Wang, Yuxiong
Department of Study
Electrical & Computer Eng
Discipline
Electrical & Computer Engr
Degree Granting Institution
University of Illinois Urbana-Champaign
Degree Name
Ph.D.
Degree Level
Dissertation
Keyword(s)
Representation learning
Spatial-temporal data
Graph modeling
Abstract
Spatial-temporal data, from human motion and urban traffic to complex transactional systems, plays a crucial role in modern predictive analytics. Effective modeling of such data is key to capturing intricate spatial and temporal dependencies, enabling accurate forecasting and decision-making across various domains. To this end, graph-based methods have emerged as a powerful approach for representing and modeling such complex relationships.However, current research in spatial-temporal modeling faces significant challenges. Many existing approaches struggle with efficiency and scalability when handling large-scale systems. Additionally, data scarcity and incompleteness often limit model performance and generalizability. In this dissertation, we present various approaches with the goal of obtaining better representations from spatial-temporal data, to address these limitations across different applications. We begin with CoMusion, a framework for stochastic human motion prediction that models motion as a graph, departing from prior latent space approaches and showcasing the benefits of graph-centric modeling for spatial-temporal data. Building on this foundation, we introduce STMAE, a self-supervised learning method using masked autoencoders to enhance various spatial-temporal models for traffic forecasting, mitigating data scarcity and incompleteness concerns. Shifting focus to scalability, we present EiFormer, an efficient architecture employing latent attention to enable large-scale spatial-temporal forecasting in transaction systems, significantly reducing computational complexity. Finally, we apply the graph concept to visual reinforcement learning with MOOSS, which models visual observations and underlying states as graphs, proposing multi-level temporal contrastive learning to improve sample efficiency. Comprehensive experiments on various benchmark datasets demonstrate the superior performance of our proposed methods over state-of-the-art baselines.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.
