Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Contrastive General Graph Matching with Adaptive Augmentation Sampling

Jianyuan Bo, Yuan Fang

TL;DR

A novel Graph-centric Contrastive framework for Graph Matching (GCGM), capitalizing on a vast pool of graph augmentations for contrastive learning, yet without needing any side information is introduced.

Abstract

Graph matching has important applications in pattern recognition and beyond. Current approaches predominantly adopt supervised learning, demanding extensive labeled data which can be limited or costly. Meanwhile, self-supervised learning methods for graph matching often require additional side information such as extra categorical information and input features, limiting their application to the general case. Moreover, designing the optimal graph augmentations for self-supervised graph matching presents another challenge to ensure robustness and efficacy. To address these issues, we introduce a novel Graph-centric Contrastive framework for Graph Matching (GCGM), capitalizing on a vast pool of graph augmentations for contrastive learning, yet without needing any side information. Given the variety of augmentation choices, we further introduce a Boosting-inspired Adaptive Augmentation Sampler (BiAS), which adaptively selects more challenging augmentations tailored for graph matching. Through various experiments, our GCGM surpasses state-of-the-art self-supervised methods across various datasets, marking a significant step toward more effective, efficient and general graph matching.

Contrastive General Graph Matching with Adaptive Augmentation Sampling

TL;DR

A novel Graph-centric Contrastive framework for Graph Matching (GCGM), capitalizing on a vast pool of graph augmentations for contrastive learning, yet without needing any side information is introduced.

Abstract

Graph matching has important applications in pattern recognition and beyond. Current approaches predominantly adopt supervised learning, demanding extensive labeled data which can be limited or costly. Meanwhile, self-supervised learning methods for graph matching often require additional side information such as extra categorical information and input features, limiting their application to the general case. Moreover, designing the optimal graph augmentations for self-supervised graph matching presents another challenge to ensure robustness and efficacy. To address these issues, we introduce a novel Graph-centric Contrastive framework for Graph Matching (GCGM), capitalizing on a vast pool of graph augmentations for contrastive learning, yet without needing any side information. Given the variety of augmentation choices, we further introduce a Boosting-inspired Adaptive Augmentation Sampler (BiAS), which adaptively selects more challenging augmentations tailored for graph matching. Through various experiments, our GCGM surpasses state-of-the-art self-supervised methods across various datasets, marking a significant step toward more effective, efficient and general graph matching.

Paper Structure

This paper contains 31 sections, 6 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deep Graph Matching
  4. Graph Contrastive Learning
  5. Proposed Method
  6. Overall Framework
  7. Pre-training
  8. Inference
  9. Preliminaries
  10. Graph Encoder
  11. Affinity Learning and GM Solver
  12. Contrastive Loss
  13. Graph Matching Loss
  14. Comprehensive Graph Augmentations
  15. Types of Augmentation
  16. ...and 16 more sections

Figures (2)

  • Figure 1: Framework of the proposed GCGM with BiAS. (a) Pre-training: Each graph $\mathcal{G}$ in the training set is augmented by sampling augmentation pairs from a large pool, guided by the BiAS strategy. (To be clear, we only show one training graph here; the same augmentation process is applied to every graph in the training set.) (b) Inference: The pre-trained model is frozen and applied to unseen graph pairs.
  • Figure 2: Graph augmentation: (a) input graph; (b) node insertion; (c) node replacement; (d) edge removal. The blue node represents the inserted node, and the dotted edge indicates the added edge.