Summary of Beyond Over-smoothing: Uncovering the Trainability Challenges in Deep Graph Neural Networks, by Jie Peng et al.
Beyond Over-smoothing: Uncovering the Trainability Challenges in Deep Graph Neural Networks
by Jie Peng, Runlin Lei, Zhewei Wei
First submitted to arxiv on: 7 Aug 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 The abstract presents research on Graph Neural Networks (GNNs) that challenges the prevailing understanding of their performance degradation when depth exceeds 8-10 layers, attributed to Over-smoothing. By analyzing this phenomenon through empirical experiments and theoretical gradient analysis, the authors identify the actual root cause as trainability issues in deep Multi-Layer Perceptrons (MLPs). They demonstrate that existing methods aimed at addressing Over-smoothing actually improve MLP trainability, leading to performance gains. Furthermore, they show that constrained gradients can enhance GNN trainability. Experimental results on diverse datasets validate these findings. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Deep Graph Neural Networks (GNNs) have a surprising problem: their performance gets worse when they get deeper. Researchers thought this was because of something called Over-smoothing, but now we know that’s not the main issue. Instead, it’s the way deep neural networks are trained that’s causing the problem. The authors of this paper studied this and found that many techniques designed to fix Over-smoothing actually help train these networks better. They also showed that by controlling how gradients change, they can make GNNs easier to train. This is important because it helps us build better models for understanding complex data like social networks. |
Keywords
* Artificial intelligence * Gnn
