UNITE:Universally Trustworthy GNN Via Subgraph Identification
Primary Area: societal considerations including fairness, safety, privacy
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: GNN, Trustworthy, Explainability, Robustness, Fairness
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Graph Neural Networks (GNNs) have become instrumental in modeling graph-structured data, with applications spanning diverse fields. Despite their prowess, challenges such as susceptibility to adversarial attacks, inherent biases, and opacity in decision-making processes have emerged. While efforts exist to address individual trustworthiness facets like robustness, interpretability, and fairness, a comprehensive solution remains elusive. This study introduces \Algname(\unite), a novel end-to-end framework uniquely designed to holistically integrate these dimensions. Unlike traditional approaches, \Algname leverages the intricate relationships between these aspects in graph data, presenting optimization goals grounded in information-theoretic principles. Preliminary experiments on real-world datasets indicate that \Algname outperforms existing methods, achieving a harmonious blend of interpretability, robustness, and fairness. This work addresses the pressing challenges in GNNs for trustworthy graph neural networks, paving the way for their broader adoption in critical domains.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 7688
Loading