SA$^{2}$GFM: Enhancing Robust Graph Foundation Models with Structure-Aware Semantic Augmentation

TL;DR

SA$^{2}$GFM introduces a structure-aware semantic augmentation framework for graph foundation models, significantly improving their robustness and domain adaptation capabilities against noise and adversarial attacks.

Contribution

The paper proposes a novel framework that encodes hierarchical structural priors, employs a self-supervised information bottleneck, and introduces expert routing for enhanced robustness and adaptation.

Findings

Outperforms 9 state-of-the-art baselines in robustness and effectiveness.

Enhances node and graph classification under noise and adversarial attacks.

Improves domain-adaptive representations through structure-aware augmentation.

Abstract

We present Graph Foundation Models (GFMs) which have made significant progress in various tasks, but their robustness against domain noise, structural perturbations, and adversarial attacks remains underexplored. A key limitation is the insufficient modeling of hierarchical structural semantics, which are crucial for generalization. In this paper, we propose SA$^{2}$GFM, a robust GFM framework that improves domain-adaptive representations through Structure-Aware Semantic Augmentation. First, we encode hierarchical structural priors by transforming entropy-based encoding trees into structure-aware textual prompts for feature augmentation. The enhanced inputs are processed by a self-supervised Information Bottleneck mechanism that distills robust, transferable representations via structure-guided compression. To address negative transfer in cross-domain adaptation, we introduce an expert adaptive routing mechanism, combining a mixture-of-experts architecture with a null expert design. For efficient downstream adaptation, we propose a fine-tuning module that optimizes hierarchical structures through joint intra- and inter-community structure learning. Extensive experiments demonstrate that SA$^{2}$GFM outperforms 9 state-of-the-art baselines in terms of effectiveness and robustness against random noise and adversarial perturbations for node and graph classification.