Summary of Deep Graph Attention Networks, by Jun Kato et al.
Deep Graph Attention Networks
by Jun Kato, Airi Mita, Keita Gobara, Akihiro Inokuchi
First submitted to arxiv on: 21 Oct 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 Graph neural networks (GNNs) often struggle with over-smoothing, where node representations become similar across different classes as layer depth increases. To address this issue, we introduce DeepGAT, a method that constructs graph attention networks (GATs) without requiring protracted tuning of the number of layers. By ensuring distinctness between node representations in different classes at each layer, DeepGAT prevents over-smoothing and achieves similar performance to a 2-layer GAT using a 15-layer network. Our approach enables training large networks with comparable attention coefficients to those with few layers, thereby saving time and enhancing GNN performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Graphs are useful for representing real-world objects. A problem with graph neural networks (GNNs) is that they tend to lose their differences as the number of layers increases. This makes it hard for them to accurately identify different classes. To solve this issue, we created a new method called DeepGAT. It helps GNNs learn and avoid losing these differences by keeping nodes in different classes distinct at each layer. With DeepGAT, we can train networks with many layers without worrying about losing their accuracy. |
Keywords
» Artificial intelligence » Attention » Gnn
