Toward Interpretable and Stable Graph Neural Networks
Abstract
As new generalizations of traditional deep neural networks to graph structured data, Graph Neural Networks (or GNNs) have demonstrated the power in graph representation learning and have permeated numerous areas of science and technology. However, GNNs also inherited drawbacks of traditional deep neural networks including lack of interpretability and vulnerable and unstable to adversarial attacks. These drawbacks have raised tremendous concerns to adopt GNNs in many critical applications pertaining to fairness, privacy, and safety. Though there are very recent efforts on the research of GNNs in terms of interpretability and vulnerability and building stable GNN models (or stability) , these studies are still at the stage of initial development and a comprehensive investigation of these new frontiers of GNNs is critically desired. This project aims to tackle the major drawbacks of GNNs and greatly enlarge their usability in critical applications. To achieve the research goal, we systematically investigate the primary directions of GNNs including new mechanisms to interpret GNNs, and ingenious strategies to attack and secure GNNs. Each direction will dramatically extend the frontier through not only studying original problems, but also developing innovative solutions. The significance of the project lies in the fact that the project offers the first comprehensive investigation on these new frontiers and the designed novel methodologies and tasks will deepen our understanding on the inner working mechanisms of GNNs and contribute to real-world applications. The success of this project will be (1) New interpretable and stable GNNs with state of the art predictive performance; (2) Theoretical analysis such as convergence and complexity; and (3) Open-source implementations of all key algorithms and frameworks. The proposed research agenda provides new perspectives for graph neural networks research, investigates original problems that entreat innovative solutions, and paves the way for new research endeavors to effectively tame graph structured data for discovering actionable patterns and harnessing them for advancing related applications. The results of this project will largely push the research boundaries of GNNs in terms of interpretability, vulnerability and stability and maximally mitigate peopleÕs concerns on adopting GNNs in their domains . Thus, it has the potential to impact the successful adoption and use of GNNs in a broad range of fields such as Computer Science, Social Science, Health Informatics, Bioinformatics and Education and Military. The proposed research will involve graduate and undergraduate students in pursuing their theses or honorÕs projects. Discoveries and research findings of this project will be tightly integrated into several current and new courses at PSU and MSU. Instructional content will be created to enable fast distribution of our results to a wide audience, and tools will be built to help data science knowledge awareness and adoption. Customized support will be developed and implemented to attract K-12 teachers and students to engage in network analytics research. The findings of this project will be timely disseminated via multiple means such as a data repository, journal and conference publications, special purpose workshops co-held at prominent conferences, and industrial participation such as internships and data sharing projects.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 25, 2021
- Source ID
- W911NF2110198
Entities
People
- Suhang Wang
Organizations
- Army Contracting Command
- Pennsylvania State University
- United States Army