![[2602.16947] Beyond Message Passing: A Symbolic Alternative for Expressive and Interpretable Graph Learning](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16947] Beyond Message Passing: A Symbolic Alternative for Expressive and Interpretable Graph Learning
Summary
The paper presents SymGraph, a novel symbolic framework that enhances graph learning by overcoming limitations of traditional message-passing methods, achieving superior expressiveness and interpretability.
Why It Matters
As Graph Neural Networks (GNNs) become critical in fields like drug discovery, their interpretability is essential for trust. SymGraph addresses the black-box nature of GNNs, offering a more transparent alternative that could lead to advancements in explainable AI and scientific discovery.
Key Takeaways
- SymGraph surpasses the 1-Weisfeiler-Lehman expressivity barrier.
- The framework achieves significant speedups in training time (10x to 100x) using only CPU execution.
- SymGraph generates rules with superior semantic granularity compared to existing methods.
- The approach enhances interpretability in high-stakes domains like drug discovery.
- Empirical evaluations show SymGraph outperforms existing self-explainable GNNs.
Computer Science > Machine Learning arXiv:2602.16947 (cs) [Submitted on 18 Feb 2026] Title:Beyond Message Passing: A Symbolic Alternative for Expressive and Interpretable Graph Learning Authors:Chuqin Geng, Li Zhang, Haolin Ye, Ziyu Zhao, Yuhe Jiang, Tara Saba, Xinyu Wang, Xujie Si View a PDF of the paper titled Beyond Message Passing: A Symbolic Alternative for Expressive and Interpretable Graph Learning, by Chuqin Geng and 7 other authors View PDF HTML (experimental) Abstract:Graph Neural Networks (GNNs) have become essential in high-stakes domains such as drug discovery, yet their black-box nature remains a significant barrier to trustworthiness. While self-explainable GNNs attempt to bridge this gap, they often rely on standard message-passing backbones that inherit fundamental limitations, including the 1-Weisfeiler-Lehman (1-WL) expressivity barrier and a lack of fine-grained interpretability. To address these challenges, we propose SymGraph, a symbolic framework designed to transcend these constraints. By replacing continuous message passing with discrete structural hashing and topological role-based aggregation, our architecture theoretically surpasses the 1-WL barrier, achieving superior expressiveness without the overhead of differentiable optimization. Extensive empirical evaluations demonstrate that SymGraph achieves state-of-the-art performance, outperforming existing self-explainable GNNs. Notably, SymGraph delivers 10x to 100x speedups in training time using...
