Out-of-Distribution Robust Explainer for Graph Neural Networks

HGOOD : Hypergraph-enhanced Graph Contrastive Learning for Graph Out-of-Distribution Detection

Conceptual Relationship Between Classical Machine Learning and Graph Neural Networks in Out-of-Distribution Detection: A Comprehensive Survey

Out-of-Distribution (OOD) detection, an important branch of Anomaly Detection (AD), aims to recognize inputs that fall outside the training distribution. Existing OOD methods based on statistical modeling, distance measures, one-class classification, reconstruction, or energy-based scoring provide strong foundations, but they can be difficult to apply directly to high-dimensional, unstructured, or relational data. Recent advances in graph-based deep learning, particularly Graph Neural Networks (GNNs), create new opportunities by modeling local and global dependencies among interdependent samples. This survey provides a comprehensive overview of OOD detection methods, organizing them into traditional and deep learning families before examining graph-aware GNN-based approaches. Beyond taxonomy, we synthesize the conceptual relationships between these paradigms through three working hypotheses: graph-aware transfer, structure-sensitive scoring, and hybrid objective design. This hypothesis-guided synthesis characterizes how established statistical, geometric, reconstruction-based, and energy-based principles can be transferred, reformulated, and combined within modern GNN frameworks. To make this perspective concrete, we include Graph Energy-based OOD Detection (GEO), a reference case study that combines energy-based scoring with one-class regularization in a GNN architecture and reports competitive but setting-dependent performance across representative benchmarks. Taken together, the survey clarifies the current landscape of OOD detection and highlights how integrating classical statistical reasoning with graph-based learning can inform future graph OOD research. To support reproducibility, the implementation of GEO is publicly available at https://github.com/longstnguyen/GEO.

Unified counterfactual explainer for graph neural networks

Counterfactual explanations emerged as a powerful tool to unveil the opaque decision-making processes of Graph Neural Networks (GNNs). However, existing techniques primarily focus on edge modifications, often overlooking the crucial role of node features and edge attributes changes in shaping the model predictions. To address this limitation, we propose a novel Unified Counterfactual Explainer (UCExplainer) for GNNs that generates counterfactual explanations for both node and graph classification tasks. Unlike prior methods, UCExplainer generates realistic and interpretable counterfactuals through a unified approach. It balances modifications by jointly optimizing perturbations across edges, edge attributes, and node features, ensuring the minimal changes required to flip a model’s prediction. Furthermore, UCExplainer offers maximum flexibility by supporting: (1) perturbations of both discrete and continuous values, (2) the clipping of values to a user-defined ranges, and (3) the exclusion of specific elements, such as entire nodes, edges, or individual features, from modification. Experiments on real-world datasets demonstrate the effectiveness and robustness of our approach over existing baselines.

ADC-GNN: Adaptive Dual-level Collaborative Graph Neural Networks for Graph Classification

Examining the Efficacy of Graph Neural Network Message-Passing in Regression Contexts

Graph Neural Networks (GNN) facilitate effective prediction on graph data such as molecules, media networks and neural network blueprints. GNNs facilitate prediction through message passing techniques which define how information flows from a node to its neighbors. Due to the ubiquity of the graph data type, the development of newer and better GNNs has garnered much interest in the machine learning community. However, GNN evaluation and benchmarking is primarily driven by classification tasks. Thus, prospective GNN message passing layers are evaluated on their ability to outperform prior work in classification contexts. In contrast, GNNs are equally capable of performing scalar regression prediction, yet this class of problem is often overlooked when proposing new GNNs while the best classification GNNs are utilized in an a priori or off-the-shelf manner for regression problems. In response, this paper studies the efficacy of GNN layers in a slew of regression contexts from rank ordering, error minimization and insight extraction. Results show that deep convolutional GNNs, particularly GEN, are more effective at these tasks than attention-based GNNs, while other classical, theoretically-inspired GNNs remain competitive and efficient.

DeltaGNN: Graph Neural Network with Information Flow Control

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.