Adaptive Prototype-Guided Personalized Propagation for Heterophilic Graphs With Missing Data.

Graph neural networks (GNNs) with missing node features have recently received increasing interest. Such missing node features seriously hurt the performance of the existing GNNs. Some recent methods have been proposed to reconstruct the missing node features by the information propagation among nodes with known and unknown attributes. Although these methods have achieved superior performance, how to exactly exploit the complex data correlations among nodes to reconstruct missing node features is still a great challenge. To solve the above problem, we propose a self-supervised guided hypergraph feature propagation (SGHFP). Specifically, the feature hypergraph is first generated according to the node features with missing information. And then, the reconstructed node features produced by the previous iteration are fed to a two-layer GNNs to construct a pseudo-label hypergraph. Before each iteration, the constructed feature hypergraph and pseudo-label hypergraph are fused effectively, which can better preserve the higher-order data correlations among nodes. After then, we apply the fused hypergraph to the feature propagation for reconstructing missing features. Finally, the reconstructed node features by multi-iteration optimization are applied to the downstream semi-supervised classification task. Extensive experiments demonstrate that the proposed SGHFP outperforms the existing semi-supervised classification with missing node feature methods.

Learning node representations against perturbations

Graph data is omnipresent and has a wide variety of applications, such as in natural science, social networks, or the semantic web. However, while being rich in information, graphs are often noisy and incomplete. As a result, graph completion tasks, such as node classification or link prediction, have gained attention. On one hand, neural methods, such as graph neural networks, have proven to be robust tools for learning rich representations of noisy graphs. On the other hand, symbolic methods enable exact reasoning on graphs. We propose Knowledge Enhanced Graph Neural Networks (KeGNN), a neuro-symbolic framework for graph completion that combines both paradigms as it allows for the integration of prior knowledge into a graph neural network model. Essentially, KeGNN consists of a graph neural network as a base upon which knowledge enhancement layers are stacked with the goal of refining predictions with respect to prior knowledge. We instantiate KeGNN in conjunction with two well-known graph neural networks, Graph Convolutional Networks and Graph Attention Networks, and evaluate KeGNN on multiple benchmark datasets for node classification.

Differentially private graph neural networks for graph classification and its adaptive optimization

Graph neural networks (GNNs) have demonstrated their effectiveness in various tasks supported by their generalization capabilities. However, the current analysis of GNN generalization relies on the assumption that training and testing data are independent and identically distributed (i.i.d). This imposes limitations on the cases where a model mismatch exists when generating testing data. In this paper, we examine GNNs that operate on geometric graphs generated from manifold models, explicitly focusing on scenarios where there is a mismatch between manifold models generating training and testing data. Our analysis reveals the robustness of the GNN generalization in the presence of such model mismatch. This indicates that GNNs trained on graphs generated from a manifold can still generalize well to unseen nodes and graphs generated from a mismatched manifold. We attribute this mismatch to both node feature perturbations and edge perturbations within the generated graph. Our findings indicate that the generalization gap decreases as the number of nodes grows in the training graph while increasing with larger manifold dimension as well as larger mismatch. Importantly, we observe a trade-off between the generalization of GNNs and the capability to discriminate high-frequency components when facing a model mismatch. The most important practical consequence of this analysis is to shed light on the filter design of generalizable GNNs robust to model mismatch. We verify our theoretical findings with experiments on multiple real-world datasets.

Graph neural networks (GNNs) have recently gained much attention for node and graph classification tasks on graph-structured data. However, multiple recent works showed that an attacker can easily make GNNs predict incorrectly via perturbing the graph structure, i.e., adding or deleting edges in the graph. We aim to defend against such attacks via developing certifiably robust GNNs. Specifically, we prove the first certified robustness guarantee of any GNN for both node and graph classifications against structural perturbation. Moreover, we show that our certified robustness guarantee is tight. Our results are based on a recently proposed technique called randomized smoothing, which we extend to graph data. We also empirically evaluate our method for both node and graph classifications on multiple GNNs and multiple benchmark datasets. For instance, on the Cora dataset, Graph Convolutional Network with our randomized smoothing can achieve a certified accuracy of 0.49 when the attacker can arbitrarily add/delete at most 15 edges in the graph.

IntroductionReconstructing low-level particle tracks in neutrino physics can address some of the most fundamental questions about the universe. However, processing petabytes of raw data using deep learning techniques poses a challenging problem in the field of High Energy Physics (HEP). In the Exa.TrkX Project, an illustrative HEP application, preprocessed simulation data is fed into a state-of-art Graph Neural Network (GNN) model, accelerated by GPUs. However, limited GPU memory often leads to Out-of-Memory (OOM) exceptions during training, due to the large size of models and datasets. This problem is exacerbated when deploying models on High-Performance Computing (HPC) systems designed for large-scale applications.MethodsWe observe a high workload imbalance issue during GNN model training caused by the irregular sizes of input graph samples in HEP datasets, contributing to OOM exceptions. We aim to scale GNNs on HPC systems, by prioritizing workload balance in graph inputs while maintaining model accuracy. Our paper introduces diverse balancing strategies aimed at decreasing the maximum GPU memory footprint and avoiding the OOM exception, across various datasets.ResultsOur experiments showcase memory reduction of up to 32.14% compared to the baseline. We also demonstrate the proposed strategies can avoid OOM in application. Additionally, we create a distributed multi-GPU implementation using these samplers to demonstrate the scalability of these techniques on the HEP dataset.DiscussionBy assessing the performance of these strategies as data loading samplers across multiple datasets, we can gauge their effectiveness in both single-GPU and distributed environments. Our experiments, conducted on datasets of varying sizes and across multiple GPUs, broaden the applicability of our work to various GNN applications that handle input datasets with irregular graph sizes.

This article introduces the Missingness-aware Self-Supervised Graph Neural Network (MissGNN), a novel model for tabular data predictions in scenarios with missing values. Addressing a significant challenge in fields like finance, healthcare, and environmental studies, MissGNN innovatively integrates feature imputation and label prediction within a unified framework. Utilizing a Tabular Bipartite Graph, it distinctively represents data points and features, enabling intricate modeling of their relationships. This approach is pivotal in contexts where missing data can significantly skew model training and predictions. MissGNN employs a Graph Neural Network to generate embeddings, capturing interactions between data points and features for accurate missing value imputation and label prediction. It features a dual-focus mechanism, comprising edge-wise feature imputation and node-wise label prediction. A pre-training phase for feature imputation enhances the model’s predictive ability for both features and labels. Additionally, MissGNN incorporates a multi-view learning strategy through feature subsetting, further enriching its representation learning. Experiments on 14 tabular datasets demonstrate MissGNN’s superiority over existing models in managing missing values across both regression and classification tasks. The results underscore its robustness and practical applicability, establishing it as a versatile tool for tackling tabular data prediction challenges with missing values. MissGNN’s contributions significantly redefine the approach to missing value imputation, aligning feature imputation and label prediction in a cohesive manner and underscoring the potential of Graph Neural Networks in processing tabular data.

Purpose Graph neural networks (GNNs) have been widely applied to various recommendation scenarios due to their outstanding performance in capturing and modelling hidden relationships. However, as the performance of a GNN heavily relies on the graph structure, the structure bias can lead to significant unfairness problems in a GNN-based recommender system across different user groups. Furthermore, prior studies have shown that sensitive attributes of nodes can be revealed through the graph structure. In other words, even if a GNN-based recommendation system does not explicitly disclose a user’s sensitive information, it can still be inferred from the imbalance in users’ connections within the graph. This article therefore addresses the unfairness arising from such structural biases by proposing a novel model designed to mitigate these biases in GNN-based recommendation. Here, the fairness requirements refer to preventing the exposure of sensitive attributes during user modeling and avoiding biased recommendations across different demographic groups. Design/methodology/approach To address this issue, we propose a novel fairness-aware graph neural network model, EqStruct-GNN, which aims to mitigate inter-group unfairness at the structural level. The model is composed of three key modules: (1) the Adaptive Behavior Modeling Module, which computes fairness-aware node behavior probability distributions; (2) the Fairness-Aware Structure Construction Module, which builds graph structures based on these distributions and (3) the Fair Representation Learning Module, which generates unbiased node embeddings. Together, these modules enable EqStruct-GNN to construct a fairness-aware graph that reduces structural bias and improves fairness across sensitive groups. Findings Experimental results demonstrate that the proposed method achieves strong recommendation performance while ensuring fairness across multiple datasets. Furthermore, when applied to various backbone models, EqStruct-GNN not only significantly improves fairness metrics but also maintains or even enhances recommendation accuracy. These results highlight the generalization capability and practical effectiveness of the proposed framework. Originality/value Unlike existing methods that focus on representation-level fairness, our model tackles structural bias at its root by building a fairness-aware debiased graph through a novel graph construction mechanism. This structure enables the learning of fair and high-quality node representations by effectively mitigating structural bias while preserving critical interaction information. In addition, the proposed approach is universally adaptable and can be seamlessly applied to various GNN architectures, demonstrating strong generality and practical applicability.

Graph Neural Networks (GNNs) have achieved remarkable success in analyzing graph-structured data, with their performance dependent on the graph structure. However, models trained on high-quality graph structures often suffer a significant performance drop when evaluated on perturbed graphs. Existing methods tackle this problem by improving the robustness of GNNs, but they often overlook representation deviation caused by structural changes. To address this limitation, we propose an attribute-guided dynamic prompt learning model that generates prompt vectors to approximate the intrinsic information of nodes. With these prompt vectors, the trained GNNs are expected to maintain their performance under perturbed graph structures. Unlike previous prompt-based methods that learn unified prompt vectors for all nodes, we obtain node-level prompts by encoding node attributes that provide unique information. Given the diversity of perturbed graph structures during inference, we introduce a structure-aware adaptation mechanism that adjusts the prompt vectors based on the input graph. Furthermore, we apply gradient-based attacks to generate perturbed graphs, encouraging the model to generalize to unseen structures. Extensive experiments on multiple benchmark datasets demonstrate the effectiveness and robustness of our model.

Knowledge Tracing (KT) aims to model students' evolving knowledge states and predict their future performance. Traditional sequential models often struggle with long-range dependencies due to inefficient memory retention and high computational costs, limiting their effectiveness in capturing students' continuous learning trajectories. Meanwhile, graph-based KT models effectively represent knowledge concept relationships but are not designed for sequential modeling. To address these challenges, we propose DualFlowKT, a novel model that employs a dual-path architecture, integrating Graph Neural Networks (GNNs) for structured knowledge modeling and the Mamba state space model for efficient long-sequence learning. Specifically, the graph-based path extracts relational dependencies among knowledge concepts, while the sequential path leverages Mamba's selective state-space modeling to effectively capture long-range learning patterns with enhanced memory efficiency and reduced computational overhead. To further enhance input representations, we introduce a Feature Enhancement Module (FEM) that extracts critical learning features such as concept continuity, learning rate, and adaptive difficulty. Additionally, a gated fusion mechanism dynamically integrates knowledge structure representations from GNNs with sequential dependencies captured by Mamba, allowing the model to adaptively balance structural and temporal learning information. Extensive experiments on multiple KT benchmark datasets demonstrate that DualFlowKT outperforms existing methods, particularly in handling long-sequence interactions and complex knowledge structures. Furthermore, the model provides deeper insights into students' evolving knowledge states, supporting personalized education systems.

Graph Neural Networks (GNNs) have demonstrated a great representation learning capability on graph data and have been utilized in various downstream applications. However, real-world data in web-based applications (e.g., recommendation and advertising) always contains bias, preventing GNNs from learning fair representations. Although many works were proposed to address the fairness issue, they suffer from the significant problem of insufficient learnable knowledge with limited attributes after debiasing. To address this problem, we develop Graph-Fairness Mixture of Experts (G-Fame), a novel plug-and-play method to assist any GNNs to learn distinguishable representations with unbiased attributes. Furthermore, based on G-Fame, we propose G-Fame++, which introduces three novel strategies to improve the representation fairness from node representations, model layer, and parameter redundancy perspectives. In particular, we first present the embedding diversified method to learn distinguishable node representations. Second, we design the layer diversified strategy to maximize the output difference of distinct model layers. Third, we introduce the expert diversified method to minimize expert parameter similarities to learn diverse and complementary representations. Extensive experiments demonstrate the superiority of G-Fame and G-Fame++ in both accuracy and fairness, compared to state-of-the-art methods across multiple graph datasets.

The use of electronic medical record (EMR) to develop automated disease prediction has been suffering from sparse features. Graph neural network (GNN) can model the non-grid data and infer missing features, which is an effective method to solve the above problem. However, the relation of two entity nodes is difficult to explored by GNN, if they cross more than three sentences, which may loss some semantic information for disease inferring. Furthermore, temporal dependency of nodes in EMRs not only provides more information for predicting disease, but also enhances ability of semantic discrimination of node representations. This study proposes a knowledge-based spatio-temporal attention network (KSAN) for disease prediction. Knowledge graph and relation completion are firstly leveraged to build more relations among nodes to supply potential disease-related facts. Secondly, the dependency among nodes are comprehensively aggregated from spatio-temporal multidimension. Meanwhile, the key nodes' weights are increased by matching a key factor set, which provides more goal-directed information. Finally, the representative embedding is generated for disease prediction. The training of KSAN is based on semi-supervised learning, which helps to save resources. KSAN is evaluated using a real-world Chinese EMR dataset, the results indicate that KSAN achieves the best prediction performance.

Drug development is a complex and lengthy process, and drug-disease association prediction aims to significantly improve research efficiency and success rates by precisely identifying potential associations. However, existing methods for drug-disease association prediction still face limitations in feature representation, feature integration, and generalization capabilities. To address these challenges, we propose a novel model named AMFGNN (Adaptive Multi-View Fusion Graph Neural Network). This model leverages an adaptive graph neural network and a graph attention network to extract drug features and disease features, respectively. These features are then used as the initial representations of nodes in the drug-disease association network to enable efficient information fusion. Additionally, the model incorporates a contrastive learning mechanism, which enhances the similarity and differentiation between drugs and diseases through cross-view contrastive learning, thereby improving the accuracy of association prediction. Furthermore, a Kolmogorov-Arnold network is employed to perform weighted fusion of various final features, optimizing prediction performance. AMFGNN demonstrates a significant advantage in predictive performance, achieving an average AUC value of 0.9453, which reflects the model's high accuracy in prediction. Cross-validation results across multiple datasets indicate that AMFGNN outperforms seven advanced drug-disease association prediction methods. Additionally, case studies on Hepatoblastoma, asthma and Alzheimer's disease further confirm the model's effectiveness and potential value in real-world applications.

Spatial domain identification, a pivotal task in spatial transcriptomics (ST) research, seeks to elucidate the spatial distribution relationships among diverse cell types and complex tissue architectures. Recent advancements in spatial transcriptomics technology have substantially improved our understanding of cellular spatial arrangements, intercellular interactions, and the local regulation of gene expression. However, conventional algorithms often face challenges in handling high-dimensional, sparse spatial transcriptomic data, which are highly susceptible to noise interference. To address these issues effectively, we propose a novel Transformer-based spatial domain recognition algorithm that synergizes the strengths of Transformer models and graph neural networks. This approach is specifically designed to overcome the limitations of high-dimensional sparse data. By representing spatial transcriptome data through a Transformer encoder structure, our model harnesses self-attention mechanisms and inter-layer dependencies to capture complex intercellular relationships. The self-attention mechanism enables global learning of dependencies across spatial regions, thereby enhancing the modeling of intricate tissue structures. Moreover, during training, we incorporate data augmentation and regularization techniques to mitigate overfitting risks and improve generalization capabilities across varied datasets. Experimental results indicate that our model achieves state-of-the-art performance on multiple datasets, significantly boosting the accuracy of spatial domain recognition and demonstrating heightened robustness in distinguishing cell types at different resolutions.