GALA: Graph Diffusion-based Alignment with Jigsaw for Source-free Domain Adaptation

TL;DR

This study proposes a novel method named Graph Diffusion-based Alignment with Jigsaw (GALA), tailored for source-free graph domain adaptation, and develops a simple yet effective graph-mixing strategy named graph jigsaw to combine confident graphs and unconfident graphs, which can enhance generalization capabilities and robustness via consistency learning.

Abstract

Source-free domain adaptation is a crucial machine learning topic, as it contains numerous applications in the real world, particularly with respect to data privacy. Existing approaches predominantly focus on Euclidean data, such as images and videos, while the exploration of non-Euclidean graph data remains scarce. Recent graph neural network (GNN) approaches can suffer from serious performance decline due to domain shift and label scarcity in source-free adaptation scenarios. In this study, we propose a novel method named Graph Diffusion-based Alignment with Jigsaw (GALA), tailored for source-free graph domain adaptation. To achieve domain alignment, GALA employs a graph diffusion model to reconstruct source-style graphs from target data. Specifically, a score-based graph diffusion model is trained using source graphs to learn the generative source styles. Then, we introduce perturbations to target graphs via a stochastic differential equation instead of sampling from a prior, followed by the reverse process to reconstruct source-style graphs. We feed the source-style graphs into an off-the-shelf GNN and introduce class-specific thresholds with curriculum learning, which can generate accurate and unbiased pseudo-labels for target graphs. Moreover, we develop a simple yet effective graph-mixing strategy named graph jigsaw to combine confident graphs and unconfident graphs, which can enhance generalization capabilities and robustness via consistency learning. Extensive experiments on benchmark datasets validate the effectiveness of GALA.

Figures (9)

• Figure 1: A brief motivation of GALA. Previous score-based diffusion models usually generate new data by sampling from a prior distribution (upper), while ours GALA transforms the target graphs back to the source domain (lower).
• Figure 2: Overview of GALA. We employ source graphs to train a score-based graph diffusion model and then transform target graphs into source-style graphs. Moreover, we introduce adaptive class-specific thresholds to generate confident graphs with pseudo-labels and then utilize graph jigsaw to exchange subgraphs between graph pairs for consistency learning.
• Figure 3: Visualization of diffusion adaptation on ENZYMES. E0 and E3 are subsets of ENZYMES, with E0 being sparser and E3 being denser. The diffusion process can reconstruct the graph while preserving semantics. As $\mathbf{E3\rightarrow E0}$, the graphs become sparser. As $\mathbf{E0\rightarrow E3}$, the graphs become denser.
• Figure 4: Visualization of diffusion adaptation on PROTEINS. P0 and P3 are subsets of PROTEINS, with P0 being sparser and P3 being denser. The diffusion process can reconstruct the graph while preserving semantics. As $\mathbf{P3\rightarrow P0}$, the graphs become sparser. As $\mathbf{P0\rightarrow P3}$, the graphs become denser.
• Figure 5: Visualization of graph jigsaw on ENZYMES. Graph jigsaw, combined with consistency learning, leverages unlabeled target data to enhance the generalization capacity.