Summary of Alleviating Over-smoothing Via Aggregation Over Compact Manifolds, by Dongzhuoran Zhou et al.
Alleviating Over-Smoothing via Aggregation over Compact Manifolds
by Dongzhuoran Zhou, Hui Yang, Bo Xiong, Yue Ma, Evgeny Kharlamov
First submitted to arxiv on: 27 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach to address the over-smoothing issue in graph neural networks (GNNs) is proposed, which replaces traditional information aggregation with aggregation over compact manifolds. The study reveals that existing techniques for combating over-smoothing rely on contracted aggregations, which inevitably lead to feature convergence. To overcome this limitation, the authors introduce the ACM method, demonstrating its effectiveness in alleviating over-smoothing and outperforming state-of-the-art models through extensive empirical evaluation. This research contributes to the development of GNNs by providing a deeper understanding of their limitations and proposing a new approach to improve their performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary GNNs have become very good at solving certain problems, but they also have some weaknesses. One big problem is that the information they learn can get mixed up and lose its unique features after many layers. Researchers have tried different ways to fix this issue, but none of them completely solve the problem. In a new study, scientists discovered that most of these fixes rely on a type of information aggregation that makes all the features become similar. To solve this problem, they proposed a new way of aggregating information called ACM. This method helps GNNs keep their unique features and perform better than previous methods. |
