Graduation Semester and Year

Fall 2024

Language

English

Document Type

Thesis

Degree Name

Master of Science in Computer Science

Department

Computer Science and Engineering

First Advisor

Dr. Nicholas R. Gans

Second Advisor

Dr. Diego Patiño

Third Advisor

Dr. William Beksi

Abstract

Our work investigates the use of Graph Neural Networks (GNNs) for modeling complex dynamic systems. Understanding dynamic systems is essential for comprehending how physical and social interactions in real-world systems evolve over time and how they can be influenced or controlled. Conventional techniques often use differential equations and numerical methods, which are computationally expensive and demand detailed knowledge of the system’s physical behavior. GNNs provide an alternative by adopting a data-driven approach that learns directly from data, making them especially valuable for systems without closed-form solutions. This work aims to harness GNNs to predict and simulate dynamic behaviors in graph-based systems, including groups of people, animals, human-robot interactions, and vehicle traffic. We have enhanced an existing mass-spring simulation framework by incorporating more sophisticated dynamics, including varied node and edge characteristics like mass, spring constants, damping coefficients, and resting lengths, to more accurately model complex physical systems. To better isolate strengths, limitations and areas for improvement, we gradually increase the complexity of the system, introducing changes one at a time. We will present key results and observations from these enhanced simulations, underscoring the effectiveness of GNNs in capturing and simulating dynamic behaviors in complex systems.

Keywords

Graph neural networks, Dynamic systems, Modeling physical systems, Graph interaction prediction, Trajectory prediction, Mass spring damper system, modeling Complex systems, Variational autoencoder

License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.