G EO XGNN (Local Geometry Improves Explanation Robustness for Graph Neural Networks), a robust explanation framework that explicitly models the directional sensitivity of structural perturbations from a geometric perspective that significantly improves explanation robustness over existing methods.
Graph neural networks learn from relational structure but can be sensitive to edge-level noise. We study a hybrid graph classifier that combines a message-passing branch (Graph Isomorphism Network, GIN) with a topological branch based on extended persistence diagrams and PersLay embeddings. Training optionally uses a paper-specific hinge penalty motivated by stable persistence-diagram representations and Lipschitz regularity, which we abbreviate as “HK-inspired.” On six TUDataset benchmarks, we compare five ablations (Full, GIN+PersLay, GIN + HK, GIN only, PersLay only) and measure robustness as accuracy drop under 10% random edge removal at test time, with persistence diagrams fixed from the original graphs; we also report targeted perturbations on MUTAG and PROTEINS and runtime on all six datasets. The regularized configurations reduce relative accuracy degradation where GIN-only is sensitive, particularly on smaller molecular and protein graphs, but do not consistently maximize clean or perturbed accuracy. On large social-network benchmarks, robustness differences are small and clean accuracy is the main differentiator. Adding PersLay to GIN can improve accuracy on several datasets, while PersLay alone performs worst. Training the full model increases per-epoch cost relative to GIN-only, with most overhead during training rather than inference. Overall, the results support a conditional empirical conclusion: combining structure, topology, and stability-oriented training can improve robustness under specified structural perturbations, with dataset-dependent accuracy–stability trade-offs rather than universal guarantees.
Jelena Losic, Charles Fanning· PLOS Complex Systems· 0 citations
Graph Neural Networks exhibit a puzzling numerical fragility under mixed-precision training, failing significantly more often than MLPs or CNNs. This failure is inherently tied to graph structure, with heterophilic graphs and high-degree nodes being particularly vulnerable. We identify the root cause as catastrophic cancellation during neighborhood aggregation. When neighboring node embeddings point in opposite directions, their sum collapses toward zero and amplifies floating-point errors by orders of magnitude. We formalize this through the cancellation ratio ?, proving that it is fundamentally governed by graph topology, including heterophily, node degree, and network depth. Consequently, we propose Aggregation-Aware Representation Learning (AARL) to learn numerically stable and cancellation-resistant representations without sacrificing expressiveness. Unlike naive approaches that enforce neighbor alignment and destroy discriminative power, AARL maintains representation diversity while ensuring numerically safe aggregation. Experiments on diverse benchmarks demonstrate that AARL substantially improves training stability under low precision while preserving or improving classification accuracy.
Jiawei Gu, Ziyue Qiao· Proceedings of the 32nd ACM...· 0 citations
Noise Corruption is introduced, a Noise Corruption-based explanation framework, which perturbs each message through matched-norm random-direction corruption while preserving the expected squared message norm, and NICE, a Noise Corruption-based explanation framework, which learns a Stochastic Restoration Boundary under NC-induced uncertainty, balancing target-prediction restoration against compactness.
Ziluowen Luo, Jun Yin, Ruochen Liu et al.· 0 citations
Instance-level explanations aim to reveal the rationale behind a model's decisions for a specific graph. Previous methods explain graph neural networks (GNNs) by selecting important edges to induce subgraphs, where edge importance is assessed by perturbing each edge and observing changes in the model predictions. However, they often neglect the synergistic effects among edges, which are crucial for accurately characterizing edge importance. To address this issue, we propose SeeExplainer, a parameter-free explainer to interpret GNNs. Specifically, we first introduce a granular-ball graph refinement mechanism that decomposes a graph into several disjoint granular-balls with no fixed size, and utilize them as nodes to construct a structural graph. This process can better capture the synergistic effects among edges. Then, we perturb nodes and edges in the structural graph to generate explanatory subgraphs based on their respective contributions. Experiments on several graph classification datasets of different networks show that SeeExplainer outperforms state-of-the-art baselines.
Jian-Cu Chen, Shuyin Xia, Guan Wang et al.· arXiv.org· 0 citations
Graph representation learning has largely focused on designing increasingly sophisticated models to transform graph topology into vector representations, or embeddings. However, the extent to which embedding quality depends on model learning, rather than on the underlying topological transformations, remains unclear. Here, we show that informative embeddings can be derived without complicated model design and gradient-based training. Propagating random features through implicit hierarchical structures induced by random walks and anonymous walks yields embeddings that capture node proximity and structural role, respectively. These two training-free embeddings preserve complementary aspects of graph organization and perform competitively with classic and recent methods across various node-, edge-, and graph-level tasks. They often require substantially less computation, resulting in a favorable quality-efficiency trade-off. Combining the two types of embeddings further improves inference quality of some tasks compared with using either embedding type alone. Our results suggest that informative graph embeddings can arise from carefully chosen topological transformations before any learning operation is applied.
Meng Qin, Jinqiang Cui, Hongwei Zheng et al.· 0 citations
This work-in-progress paper proposes graph edit paths as a controlled tool for probing how GNN predictions respond to structural and attribute-level edits, and develops heuristics for generating edit paths and evaluates standard GNNs along these paths, quantifying their sensitivity to different types of edits.
Florian Seiffarth· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
