[공지사항]을 빙자한 안부와 근황 
Show more

Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

SamGoG: A Sampling-Based Graph-of-Graphs Framework for Imbalanced Graph Classification

Created by
  • Haebom

Author

Shangyou Wang, Zezhong Ding, Xike Xie

Outline

In this paper, we propose a sampling-based graph-of-graphs (GoG) learning framework, SamGoG, to address two important imbalance problems in graph classification tasks: class imbalance and graph size imbalance. SamGoG constructs multiple GoGs and trains them sequentially via an efficient importance-based sampling mechanism. This sampling mechanism integrates learnable pairwise similarity and adaptive GoG node degree to enhance edge homogeneity and thus improve the quality of downstream models. SamGoG can be seamlessly integrated with various downstream GNNs, enabling efficient adaptation to graph classification tasks. Extensive experiments on benchmark datasets demonstrate that SamGoG achieves state-of-the-art performance, achieving up to 15.66% accuracy improvement and 6.7x learning speedup.

Takeaways, Limitations

•
Takeaways:
◦
A new framework that effectively solves both class imbalance and graph size imbalance problems simultaneously is presented.
◦
Reduce computational costs through an efficient importance-based sampling mechanism.
◦
Improve edge homogeneity and model performance by leveraging learnable pairwise similarity and adaptive GoG node degree.
◦
Provides wide usability through compatibility with various downstream GNNs
◦
Achieves accuracy improvement of up to 15.66% and learning speed improvement of 6.7 times compared to existing methods
•
Limitations:
◦
Further research is needed on the generalization performance of the proposed method.
◦
Performance evaluation for specific types of graph data is required, and robustness verification for various graph structures is required.
◦
Further research is needed on parameter optimization of the sampling mechanism.
◦
Further review is needed for applicability and scalability to large-scale graph datasets.
👍
Made with Slashpage