FL-GNN: A Fuzzy-logic Graph Neural Network

Boyu Du, Jingya Zhou, Ling Liu, Xiaolong She

Primary Area: general machine learning (i.e., none of the above)

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: fuzzy neural network, fuzzy system, fuzzy-logic Graph Neural Network, graph inference

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Abstract: This paper presents a novel hybrid fuzzy-logic Graph Neural Network (FL-GNN) by combining Fuzzy Neural network (FNN) with GNN (Graph Neural Network) to effectively capture and aggregate local information flows within graph structural data. FL-GNN by design has three novel features. First, we introduce a specific structure fuzzy rule to boost the graph inference capability of FL-GNN to be on par with the representative GNN models. Second, we enhance the interpretability of FL-GNN by adding the analytic exploration methods to its graph inference ability from two perspectives: Fuzzy Inference System and Message Passing Algorithm (MPA). Finally, we ameliorate the structure of FL-GNN based on MPA to address the inherent limitations of FL-GNN. This optimization can reduce the calculation complexity of FL-GNN and further improve its learning efficiency. Extensive experiments are conducted to validate the graph inference capability of FL-GNN and report the performance comparison against other widely used GNN models. The results demonstrate that FL-GNN can outperform existing representative graph neural networks for graph inference tasks.

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: 5146