Summary of Page: Parametric Generative Explainer For Graph Neural Network, by Yang Qiu and Wei Liu and Jun Wang and Ruixuan Li
PAGE: Parametric Generative Explainer for Graph Neural Network
by Yang Qiu, Wei Liu, Jun Wang, Ruixuan Li
First submitted to arxiv on: 26 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 This paper presents PAGE, a novel framework for generating faithful explanations for graph neural networks without requiring prior knowledge or internal details. The authors design an auto-encoder training strategy that enables the generation of explanatory substructures. By reducing dimensionality and extracting causal features from the latent space, PAGE can explain model outputs. A discriminator is introduced to capture causality between features and output, constraining the encoder for enhanced feature generation. The framework maps these features back to graph substructures for explanations. Compared to existing methods, PAGE operates at the sample scale, eliminating the need for perturbation or encoding processes. Experimental results on artificial and real-world datasets demonstrate PAGE’s high faithfulness, accuracy, and efficiency. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary PAGE is a new way to explain how graph neural networks work without knowing their secrets. It uses an auto-encoder to create explanations that are close to reality. The authors designed a special training strategy to make this happen. By reducing the amount of information needed to understand what’s going on, PAGE can extract important features and use them to create explanations. A new way to check if these explanations are correct is also introduced. Tests show that PAGE works better than other methods in explaining things efficiently. |
Keywords
» Artificial intelligence » Encoder » Latent space
