Towards Effective and General Graph Unlearning via Mutual Evolution

TL;DR

This work tackles the challenge of graph unlearning in privacy-preserving machine learning by proposing MEGU, a mutual-evolution framework that jointly optimizes a predictive module and a forgetting unlearning module within a topology-guided objective. The approach introduces adaptive high-influence neighborhood selection and topology-aware unlearning propagation to bridge prediction and forgetting, achieving state-of-the-art results across nine graph benchmarks with notable improvements and substantial training-efficiency gains. MEGU is designed to be model-agnostic and scalable, preserving predictive performance for non-unlearning entities while efficiently forgetting unlearning elements. The framework's practical impact lies in enabling flexible, efficient, and generalizable graph unlearning suitable for real-world privacy and robustness requirements.

Abstract

With the rapid advancement of AI applications, the growing needs for data privacy and model robustness have highlighted the importance of machine unlearning, especially in thriving graph-based scenarios. However, most existing graph unlearning strategies primarily rely on well-designed architectures or manual process, rendering them less user-friendly and posing challenges in terms of deployment efficiency. Furthermore, striking a balance between unlearning performance and framework generalization is also a pivotal concern. To address the above issues, we propose \underline{\textbf{M}}utual \underline{\textbf{E}}volution \underline{\textbf{G}}raph \underline{\textbf{U}}nlearning (MEGU), a new mutual evolution paradigm that simultaneously evolves the predictive and unlearning capacities of graph unlearning. By incorporating aforementioned two components, MEGU ensures complementary optimization in a unified training framework that aligns with the prediction and unlearning requirements. Extensive experiments on 9 graph benchmark datasets demonstrate the superior performance of MEGU in addressing unlearning requirements at the feature, node, and edge levels. Specifically, MEGU achieves average performance improvements of 2.7\%, 2.5\%, and 3.2\% across these three levels of unlearning tasks when compared to state-of-the-art baselines. Furthermore, MEGU exhibits satisfactory training efficiency, reducing time and space overhead by an average of 159.8x and 9.6x, respectively, in comparison to retraining GNN from scratch.

Figures (6)

• Figure 1: Overview of our proposed MEGU. Unlearning Prediction represents the prediction of non-unlearning entities.
• Figure 2: Edge Attack performance on Cora. The x-axis is the ratio of noisy edges to the existing edges.
• Figure 3: The training trajectories of MEGU and its variants without the mutual evolution design on the same loss landscape.
• Figure 4: Sensitivity analysis on GAT backbone.
• Figure 5: Performance of different feature mask ratios on Photo and Computers with GCN and GAT backbone.

Theorems & Definitions (1)

• Definition 1