[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.

Multi-View Node Pruning for Accurate Graph Representation

Created by
  • Haebom

Author

Hanjin Kim, Jiseong Park, Seojin Kim, Jueun Choi, Doheon Lee, Sung Ju Hwang

Outline

In this paper, we propose a new graph pruning method, Multi-View Pruning (MVP), which considers the importance of nodes from multiple perspectives (multi-views) rather than simply the degree when removing nodes during graph pooling. MVP generates multiple graph views and learns the score of each node by considering both the reconstruction loss and the task loss. We experimentally demonstrate that it improves performance by combining it with existing graph pooling methods on various benchmark datasets, and that multi-view encoding and consideration of the reconstruction loss are the keys to the performance improvement.

Takeaways, Limitations

•
Takeaways:
◦
In graph pooling, we achieve performance improvement over existing methods by evaluating the importance of nodes from various perspectives.
◦
We present a novel method to effectively remove unimportant nodes by exploiting reconstruction loss.
◦
It is compatible with various graph pooling methods, showing wide applicability.
◦
Demonstrates the ability to identify low-importance nodes that match domain knowledge.
•
Limitations:
◦
There is a possibility that the performance improvement of the proposed MVP may be biased towards specific datasets or graph pooling methods.
◦
Further research is needed on optimization and generalization of various view generation methods.
◦
A more in-depth comparative analysis with other graph pruning methods may be needed.
◦
Verification of efficiency and scalability is required when applied to actual large-scale graphs.
👍
Made with Slashpage