Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Collaborate to Adapt: Source-Free Graph Domain Adaptation via Bi-directional Adaptation

Zhen Zhang, Meihan Liu, Anhui Wang, Hongyang Chen, Zhao Li, Jiajun Bu, Bingsheng He

TL;DR

This paper presents a novel paradigm called GraphCTA, which performs model adaptation and graph adaptation collaboratively through a series of procedures, and demonstrates that the proposed model outperforms recent source-free baselines by large margins.

Abstract

Unsupervised Graph Domain Adaptation (UGDA) has emerged as a practical solution to transfer knowledge from a label-rich source graph to a completely unlabelled target graph. However, most methods require a labelled source graph to provide supervision signals, which might not be accessible in the real-world settings due to regulations and privacy concerns. In this paper, we explore the scenario of source-free unsupervised graph domain adaptation, which tries to address the domain adaptation problem without accessing the labelled source graph. Specifically, we present a novel paradigm called GraphCTA, which performs model adaptation and graph adaptation collaboratively through a series of procedures: (1) conduct model adaptation based on node's neighborhood predictions in target graph considering both local and global information; (2) perform graph adaptation by updating graph structure and node attributes via neighborhood contrastive learning; and (3) the updated graph serves as an input to facilitate the subsequent iteration of model adaptation, thereby establishing a collaborative loop between model adaptation and graph adaptation. Comprehensive experiments are conducted on various public datasets. The experimental results demonstrate that our proposed model outperforms recent source-free baselines by large margins.

Collaborate to Adapt: Source-Free Graph Domain Adaptation via Bi-directional Adaptation

TL;DR

This paper presents a novel paradigm called GraphCTA, which performs model adaptation and graph adaptation collaboratively through a series of procedures, and demonstrates that the proposed model outperforms recent source-free baselines by large margins.

Abstract

Unsupervised Graph Domain Adaptation (UGDA) has emerged as a practical solution to transfer knowledge from a label-rich source graph to a completely unlabelled target graph. However, most methods require a labelled source graph to provide supervision signals, which might not be accessible in the real-world settings due to regulations and privacy concerns. In this paper, we explore the scenario of source-free unsupervised graph domain adaptation, which tries to address the domain adaptation problem without accessing the labelled source graph. Specifically, we present a novel paradigm called GraphCTA, which performs model adaptation and graph adaptation collaboratively through a series of procedures: (1) conduct model adaptation based on node's neighborhood predictions in target graph considering both local and global information; (2) perform graph adaptation by updating graph structure and node attributes via neighborhood contrastive learning; and (3) the updated graph serves as an input to facilitate the subsequent iteration of model adaptation, thereby establishing a collaborative loop between model adaptation and graph adaptation. Comprehensive experiments are conducted on various public datasets. The experimental results demonstrate that our proposed model outperforms recent source-free baselines by large margins.
Paper Structure (21 sections, 2 theorems, 17 equations, 4 figures, 8 tables, 1 algorithm)

This paper contains 21 sections, 2 theorems, 17 equations, 4 figures, 8 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. The Proposed GraphCTA
  4. Preliminary and Problem Definition
  5. Model Adaptation with Local-Global Consistency
  6. Graph Adaptation with Self-training
  7. Complexity Analysis
  8. Experiments
  9. Experimental Settings
  10. Results and Analyses
  11. Ablation Study
  12. The Effect of Model Adaptation and Graph Adaptation.
  13. The Alternative Graph Adaptation Strategies
  14. Architectures and Hyper-parameter Analyses
  15. Visualization
  16. ...and 6 more sections

Key Result

theorem 1

Given two domain distributions $\mathbb{D}_{\rm S}$ and $\mathbb{D}_{\rm T}$, denote $f^{*} = \arg \min_{f \in \mathcal{H}}(\epsilon_{\rm T}(f)+\epsilon_{\rm S}(f))$ and $\xi=\epsilon_{\rm T}(f^{*}) + \epsilon_{\rm S}(f^{*})$. Assume all hypotheses $h$ are $K$-Lipschitz continuous, the risk of hypot where $\mathcal{W}_1$ distance is used and we ignore the subscript 1 for simplicity.

Figures (4)

  • Figure 1: The overall architecture of our proposed GraphCTA framework, which consists of model adaptation and graph adaptation.
  • Figure 2: Visualizations of target graph node representations with each color representing a class in citation networks (C$\rightarrow$D).
  • Figure 3: Hyper-parameter sensitivity analysis.
  • Figure 4: The comparison of learning curves between GraphCTA and SOGA.

Theorems & Definitions (4)

  • definition 1: Source-Free Unsupervised Graph Domain Adaptation
  • definition 2: Wasserstein Distance
  • theorem 1
  • theorem 2