Stanford CS224W: Machine Learning w/ Graphs I 2023 I Label Propagation on Graphs  Summary and Q&A
TL;DR
Label propagation and correct and smooth are two methods for node classification in graphs. Label propagation uses graph structure to propagate labels and make predictions, while correct and smooth combines graph neural networks with label propagation. Both methods have been successful in various applications.
Questions & Answers
Q: What is the main difference between label propagation and correct and smooth?
The main difference is that label propagation focuses on directly propagating labels in the graph, while correct and smooth combines graph neural networks with label propagation to leverage both graph structure and node features for improved predictions.
Q: How does label propagation work?
Label propagation iteratively updates the label of each node in the graph by averaging the labels of its neighbors. This process continues until convergence or a maximum number of iterations is reached.
Q: What is the motivation behind correct and smooth?
Correct and smooth aims to overcome the limitations of graph neural networks by incorporating label propagation. It uses the idea that labels are correlated in the graph and leverages this correlation to improve predictions.
Q: Can correct and smooth be used in a graph classification problem?
Correct and smooth is primarily designed for node classification problems. However, similar techniques can be applied to graph classification by using a different formulation that incorporates label information in message passing.
Summary & Key Takeaways

Label propagation is a method that uses the graph structure to propagate labels and make predictions for unlabeled nodes in a graph.

Correct and smooth combines graph neural networks with label propagation to improve predictions by leveraging graph structure and node features.

Both methods have been successful in applications such as document classification, social network analysis, and fraud detection.