Summary of Contrastive Graph Representation Learning with Adversarial Cross-view Reconstruction and Information Bottleneck, by Yuntao Shou et al.
Contrastive Graph Representation Learning with Adversarial Cross-view Reconstruction and Information Bottleneck
by Yuntao Shou, Haozhi Lan, Xiangyong Cao
First submitted to arxiv on: 1 Aug 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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) have gained popularity for their information aggregation capabilities, but they often struggle with popularity bias in graphs and incorrect node labels. Graph Contrastive Learning (GCL) has shown promise in addressing these issues for node classification tasks. Existing GCL methods create multiple contrasting views by randomly removing edges and nodes, then maximize mutual information between them to improve node feature representation. However, this approach may learn redundant information irrelevant to the task. To address this, we propose Contrastive Graph Representation Learning with Adversarial Cross-view Reconstruction and Information Bottleneck (CGRL) for node classification. CGRL adaptively masks nodes and edges to obtain optimal graph structure representations, removes redundant information using information bottleneck theory, and adds noise perturbations to improve robustness. Our method significantly outperforms existing state-of-the-art algorithms on real-world public datasets. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Graph Neural Networks are used to analyze data that’s connected in a specific way. However, these networks often get stuck because of popular categories or wrong labels. To fix this, researchers use something called Graph Contrastive Learning. It creates many versions of the graph and makes them similar, but not exactly the same. This helps the network learn better representations of the data. We came up with a new way to do this called CGRL. It’s like a filter that removes unnecessary information and adds some noise to make it more robust. Our method works really well on real-world datasets and beats other state-of-the-art algorithms. |
Keywords
» Artificial intelligence » Classification » Representation learning
