gated multi-graph fusion

Existing approaches to jointly modeling multi-view graph structures and covariates often neglect inter-view dependencies or inadequately integrate covariate information, limiting inference accuracy. This work proposes the first unified hierarchical Bayesian framework that simultaneously models multi-view graphs—defined over a shared node set and accommodating binary or continuous edge weights—together with vector-valued covariates. By designing tailored priors, the framework enables principled graph fusion and parameter estimation while providing full quantification of uncertainty. Theoretical analysis establishes asymptotic consistency of the posterior predictive density. Empirical evaluations demonstrate superior performance over state-of-the-art methods in simulations, and a successful application in neuroscience reveals meaningful associations between brain functional connectivity during cognitive tasks and phenotypic measures.

This work addresses the challenge of joint reasoning across heterogeneous graphs in the absence of shared node identities by proposing the Multi-Graph Meta Transformer (MGMT) framework. MGMT maps individual graphs into a shared latent space using graph transformers and employs an attention mechanism to select task-relevant hypernodes. Cross-graph meta-graphs are then constructed based on latent similarities to enable joint inference. Notably, MGMT introduces an interpretable meta-graph structure that explicitly reveals cross-graph alignment relationships and critical substructures through hypernodes and hyperedges, enhancing scientific interpretability without compromising performance. Experiments demonstrate that the method outperforms state-of-the-art models on both synthetic benchmarks and real-world neuroscience tasks, while providing interpretable representations that support scientific discovery.

To address the limited modeling capacity of Graph Neural Networks (GNNs) on heterogeneous graphs, this paper proposes Hetero-GAT+LPE, a heterogeneous graph attention network augmented with full-spectrum Laplacian positional encoding. Our method is the first to incorporate full-spectrum Laplacian eigenvectors as positional encodings into heterogeneous graph attention mechanisms, jointly capturing both absolute and relative structural positions of nodes—thereby mitigating the insufficient coupling between semantic and topological information inherent in heterogeneous graphs. Extensive experiments on multiple standard heterogeneous graph benchmarks demonstrate that Hetero-GAT+LPE consistently outperforms state-of-the-art GNNs on node classification and link prediction tasks, achieving average accuracy gains of 3.2%–5.7%. These results empirically validate the critical contribution of structural positional priors to representation learning on heterogeneous graphs.

This work addresses the limitation of existing graph learning approaches, which typically operate in isolation within a single modality and task, thereby hindering the cross-task and cross-modal reuse of structural knowledge. To overcome this, the authors propose G-Substrate, a novel framework that models graph structures as persistent, shareable substrates. By unifying structural patterns and employing a role-interleaved training strategy, G-Substrate enables collaborative learning across multiple tasks and modalities. This approach facilitates the continuous accumulation and transfer of graph-structured knowledge, consistently outperforming both isolated training and conventional multi-task learning methods across diverse domains, modalities, and tasks.

Existing graph Transformers suffer from limited effectiveness, poor scalability, and high preprocessing complexity, often failing to outperform simple GNNs. To address this, we propose the first pure-attention graph learning framework that treats edge sets—not nodes—as the fundamental modeling unit, eliminating conventional node-centric representations and hand-crafted message passing. Our method introduces vertically interleaved masked and standard self-attention encoders, coupled with attention-based pooling for end-to-end differentiable training. It requires no graph reconstruction or preprocessing, natively supports heterogeneous graphs and transfer learning. Evaluated across 70+ node- and graph-level benchmark tasks, our approach consistently surpasses tuned GNN baselines and state-of-the-art graph Transformers. It achieves new SOTA results on molecular graph classification, vision-based graph recognition, heterogeneous graph learning, and cross-domain transfer, while maintaining both high accuracy and linear scalability.

Graph pre-training has achieved remarkable success in recent years, delivering transferable representations for downstream adaptation. However, most existing methods are designed for either homogeneous or heterogeneous graphs, thereby hindering unified graph modeling across diverse graph types. This separation contradicts real-world applications, where mixed homogeneous and heterogeneous graphs are ubiquitous, and distribution shifts between upstream pre-training and downstream deployment are common. In this paper, we empirically demonstrate that a balanced mixture of homogeneous and heterogeneous graph pre-training benefits downstream tasks and propose a unified multi-domain \textbf{G}raph \textbf{P}re-training method across \textbf{H}omogeneous and \textbf{H}eterogeneous graphs ($\mathbf{GPH^{2}}$). To address the lack of a unified encoder for homogeneous and heterogeneous graphs, we propose a Unified Multi-View Graph Construction that simultaneously encodes both without explicit graph-type-specific designs. To cope with the increased cross-domain distribution discrepancies arising from mixed graphs, we introduce domain-specific expert encoding. Each expert is independently pre-trained on a single graph to capture domain-specific knowledge, thereby shielding the pre-training encoder from the adverse effects of cross-domain discrepancies. For downstream tasks, we further design a Task-oriented Expert Fusion Strategy that adaptively integrates multiple experts based on their discriminative strengths. Extensive experiments on mixed graphs demonstrate that $\text{GPH}^{2}$ enables stable transfer across graph types and domains, significantly outperforming existing graph pre-training methods.

Unlike vision and language domains, graph learning lacks a shared input space, as input features differ across graph datasets not only in semantics, but also in value ranges and dimensionality. This misalignment prevents graph models from generalizing across datasets, limiting their use as foundation models. In this work, we propose ALL-IN, a simple and theoretically grounded method that enables transferability across datasets with different input features. Our approach projects node features into a shared random space and constructs representations via covariance-based statistics, thus eliminating dependence on the original feature space. We show that the computed node-covariance operators and the resulting node representations are invariant in distribution to permutations of the input features. We further demonstrate that the expected operator exhibits invariance to general orthogonal transformations of the input features. Empirically, ALL-IN achieves strong performance across diverse node- and graph-level tasks on unseen datasets with new input features, without requiring architecture changes or retraining. These results point to a promising direction for input-agnostic, transferable graph models.

This work addresses the challenge in graph foundation models where heterogeneous node features across domains are difficult to unify, and naive dimension alignment often discards critical semantic information, limiting transferability. To overcome this, the authors propose SliGFM, a novel framework that establishes four guiding principles for graph feature unification: formal consistency, cross-domain transferability, information preservation, and backbone compatibility. Guided by these principles, SliGFM introduces a topology-aware sliding-window Transformer architecture that transforms heterogeneous features into ordered, fixed-dimensional semantic tokens through topological smoothness ranking, a shared sliding-window encoder, smoothness-aware attention mechanisms, and a generative reconstruction objective. Experiments demonstrate that SliGFM effectively captures transferable relational patterns while preserving original feature semantics, significantly enhancing the generalization performance of graph foundation models across diverse downstream tasks.

本文提出PreGS框架，通过参数转移和多专家图神经网络解决单一聚合机制不足及多结构分支训练开销问题，提升节点分类性能。

Existing general-purpose graph pre-training methods struggle to effectively capture the semantic heterogeneity of node and relation types and the structural diversity of meta-paths in heterogeneous graphs, limiting their cross-dataset generalization capability. To address this, this work proposes MUG—the first universal pre-training framework tailored for heterogeneous graphs. MUG integrates multi-type information through a unified input module and employs a dimension-aware encoder to map heterogeneous features into a shared semantic space. It further introduces a meta-path-aware shared encoding mechanism to learn consistent structural patterns across diverse meta-path views, coupled with a global discriminative pre-training objective to align representations across graphs. Extensive experiments demonstrate that MUG significantly enhances cross-dataset generalization performance on multiple real-world heterogeneous graph benchmarks.