Aug 2026· Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2· 0 citations· 15 references
TL;DR
SGS-GNN improves F1-scores by 4% relative to full training and up to 30% on heterophilic graphs and outperforms state-of-the-art methods by 4–7% at similar sparsity levels while reducing peak memory usage by up to 3.9×.
Abstract
We propose SGS-GNN, a supervised graph sparsifier for Graph Neural Networks (GNNs) to improve predictive performance and reduce the cost of message passing by removing task-irrelevant edges. Existing unsupervised sparsifiers are not task-aware, while existing supervised sparsifiers suffer from significant memory overhead, poor sparsity control, and a lack of homophily/heterophily awareness. SGS-GNN addresses these limitations by adopting a feature- and structure-aware edge-probability encoder, a sparse subgraph sampler that strictly adheres to a global sparsity constraint, and a homophily-aware regularizer to improve prediction accuracy across homophilic and heterophilic graphs. A key scalability-enhancing feature of SGS-GNN is that it ensures encoder updates are computed by backpropagating through the sampled subgraph, and avoids retaining edge-level computation graphs for all edges via gradient checkpointing. A key efficiency-enhancing feature of SGS-GNN is that the edge-probability encoder is updated only when it outperforms a degree-based edge sampler, ensuring performance no worse than a strong unsupervised baseline. Experiments on 33 homophilic and heterophilic graphs show that SGS-GNN improves F1-scores by 4% relative to full training and up to 30% on heterophilic graphs. Furthermore, SGS-GNN outperforms state-of-the-art methods by 4–7% at similar sparsity levels while reducing peak memory usage by up to 3.9×.
Graph neural networks (GNNs) can exhibit unfair behavior even when sensitive attributes are excluded from node features, because graph topology and message passing propagate group-correlated signals under sensitive homophily. Existing fairness-aware GNN methods mainly constrain representations or prediction distributions at a global level, without explicitly controlling the local structural pathways through which biased information propagates during aggregation. We propose Subgraph Filtering for Fair Graph Neural Networks (SF-GNN), a lightweight and architecture-agnostic framework that mitigates structural bias at its source. SF-GNN identifies bias-prone edges by combining sensitive homophily with structural propagation amplifiers, including hub participation and triadic closure. It then incorporates stochastic edge filtering into each message-passing step to selectively downweight or remove these edges while preserving the remaining graph structure. Training further incorporates a statistical-parity regularizer with a warm-up schedule to stabilize optimization. Experiments on five benchmark datasets show that SF-GNN achieves consistent fairness improvements while maintaining competitive predictive performance, leading to a better fairness--accuracy trade-off than recent fairness-aware GNN baselines.
Haohui Lu, Ji-Yuan Tian, Fangyu Zhou et al.· 0 citations
CoRe-GNN is proposed, which performs both propagations in parallel at each layer: a coarsened inter-cluster term capturing long-range structure, and a local intra-cluster term preserving per-node discriminability.
Antonin Joly, Nicolas Keriven, Aline Roumy· 0 citations
This work proposes a novel Adaptive Dual-level Collaborative GNN associated with an adaptive dual-level collaborative mechanism, and shows that the ADC-GNN can inject the learned high-level information back into the node level, forming a closed-loop, bidirectional optimization process.
SmartNIC-based in-network prediction is a practical complement to partitioning and compression techniques for communicationefficient full-graph GNN training at scale, and results indicate SmartNIC-based in-network prediction is a practical complement to partitioning and compression techniques for communicationefficient full-graph GNN training at scale.
Guofan Yu, Sitian Chen, Zhenheng Tang et al.· IEEE International Conferenc...· 0 citations
Graph Neural Networks (GNNs) are popular deep learning models designed to process graph-structured data through recursive neighborhood aggregations in the message passing process. When applied to semi-supervised node classification, the message-passing enables GNNs to understand short-range spatial interactions, but also causes them to suffer from over-smoothing and over-squashing. These challenges hinder model expressiveness and prevent the use of deeper models to capture long-range node interactions (LRIs) within the graph. Popular solutions for LRIs detection are either too expensive to process large graphs due to high time complexity or fail to generalize across diverse graph structures. To address these limitations, we propose a mechanism called information flow control, which leverages a novel connectivity measure, called information flow score, to address over-smoothing and over-squashing with linear computational overhead, supported by theoretical evidence. Building on this mechanism, we introduce DeltaGNN, to the best of our knowledge among the first scalable (featuring linear computational and memory complexity overhead) and generalizable (capable of effectively handling graphs with diverse homophily, density, and topology) architectures for long-range and short-range interaction detection. We benchmark our model across 10 real-world datasets, including graphs with varying sizes, topologies, densities, and homophilic ratios, showing superior performance with limited computational complexity.
Kevin Mancini, Islem Rekik· IEEE Transactions on Pattern...· 2 citations