Tag
The paper introduces CIRCE, which uses a cheap LLM to extract a problem's constraint structure into a constraint graph that predicts LLM response correctness before inference, improving NLL and AUROC over existing baselines across seven reasoning benchmarks.
An arXiv paper formulates payroll fraud detection as inductive link prediction on synthetic HR/ERP graphs, comparing analyst rules, path features, and GNNs. A gradient-boosted model on chained-join features outperformed a GraIL-style GNN on link AP, while the relational entity GNN led under failed identifier joins, showing multi-hop gains are not GNN-specific.
The paper develops a degree-resolved analysis of nonlinear interactions between irreducible representations in graph neural networks, introducing Specht Triple, a residual architecture for unordered graph triples that cuts test MSE by 11.05% over a parameter-matched control on a synthetic benchmark with a frozen GIN backbone.
Caltech researchers introduce eSpinor, an SU(2)-equivariant graph neural network that propagates spinor features to predict spin-orbit-coupled magnetic properties of molecules, improving magnetic anisotropy and excitation-energy predictions for single-molecule magnets and MRI probes while generalizing to unseen ligands.
CHORD提出了一种跨模型幻觉检测（CMHD）框架，将LLM隐藏状态表示为层级-词元-特征关系图，通过关系感知注意力与元学习获得可迁移到未见目标模型的检测信号，在跨模型泛化上优于现有检测方法。
提出了 BehaveRec 三信号框架（因果会话 Transformer、90 天用户历史编码器、基于行为嵌入相似度的用户图 GraphSAGE），在 137M 条工业序列上研究长期行为历史何时真正有益于序列推荐，并得出约 20 次交互的实用部署阈值。
该论文提出 Graph-Dependent Pooling (GDPool)，一种动态决定每个图池化粒度的层级图神经网络池化方法，通过自注意力选择锚点并在金属有机框架和分子图基准上取得最佳性能。
CFBrainAug 提出一种模块度约束的反事实脑网络增强框架，将数据增强与反事实解释统一起来，在 ABIDE I 数据集上以轻量级 SVM 实现了具有竞争力的自闭症诊断精度与内在可解释性。
Delft University of Technology researchers present the first scalable framework for structural adversarial attacks and adversarial training on multigraph neural networks used for anti-money laundering detection, introducing a cycle-based regularizer to keep perturbations plausible and showing that adversarial training restores detection robustness.
该论文提出一种融合空间启发式策略的扩散强化学习方法用于在线3D装箱问题，通过启发式动作预筛、图神经网络建模空间结构以及包含平整度的复合奖励函数，提升了布局优化能力与泛化性。
This paper introduces GREG, a Graph Reasoning Model that augments Language Reasoning Models with learnable, permutation-equivariant graph positional encodings computed via a GNN and Graph Transformer, eliminating the need for serialized graph descriptions and reducing context complexity to O(N). Trained on small graphs, it achieves size-generalization to graphs with up to 1,000 nodes while remaining robust to node relabelings.
This paper frames Brick schema model generation for buildings as a knowledge graph completion problem — a first — comparing gradient-boosted trees, Pairwise Brick Relation Ranking, and CompGCN on relations like feeds, hasPart, and hasPoint. CompGCN leads on held-out buildings, but no method generalizes uniformly across datasets and relations.
RGH-GNN 提出一种风险门控的异构图神经网络，将 LightGBM 生成的交易级风险分数作为边权重引导消息传递，以提升金融欺诈检测性能，在 IEEE-CIS 数据集上取得 0.914 的 AUC-ROC 并超越所有基线。
The paper proposes SPEM, a structure-aware procedural memory framework for LLM agents that combines atomic skill memory with Skill-DAG templates and Skill-GNN retrieval, achieving 87.1% success on ALFWorld and 38.38% reward on SkillsBench while reducing token and runtime costs.
HierDTA 是一个层次结构感知的药物-靶点亲和力预测框架，将药物表示为原子-基团异构图并结合蛋白质残基交互图与交叉注意力机制，在 Davis 和 KIBA 数据集上取得领先性能，并在冷启动场景下展现出良好泛化能力。
该论文证明多项式图滤波器（线性消息传递与谱图网络的核心）在长程伊辛模型基准上存在无法通过提高阶数消除的误差下界，并提出分层步长滤波器与膨胀卷积网络，显著超越已发表的最佳结果。
该论文提出 GCRN+（图卷积循环网络），利用稀疏固定传感器（仅覆盖 15.6% 路段）结合历史探针车数据监督，在达拉斯-沃思堡高速网络上实现全网当前及未来一小时车速的估计与预测，性能显著优于历史均值及 STID、STAEformer 等基线模型。
该论文提出一种事后诊断方法：冻结已训练的GNN并在每层拟合线性探针，通过层间探针熵轨迹区分有益、中性与有害平滑，并分离异质性结构失配与过度深度过平滑两类失效模式。
O-kRystal, a periodic E(3)-equivariant graph neural network, learns corrections to electronic states beyond DFT, achieving HSE06 band predictions across nearly 6,000 materials, G0W0 quasiparticle energies in 2D systems, and a 68.7% reduction in band energy MAE on strongly correlated materials.
论文提出 FireGraphNet，将12小时野火蔓延预测建模为有向 H3 图上的节点分类任务，在加州测试火灾上平均精度达到最强距离基线的1.7–1.9倍，并分析了图消息传递对预测能力的贡献。