Skip to content

Singular Value Decomposition for Prevention of the Graph Convolutional Network Overfitting

Notifications You must be signed in to change notification settings

flaneur-ml/SVD4GCN

Repository files navigation

Singular Value Decomposition for prevention of the Graph Convolutional Network overfitting

Abstract

Our project is an effort to prevent the graph neural network from being overfitted. Graph convolutional networks are usually shallow, with the number of layers not larger than 2. Deep graph networks perform much worse, even if some standard techniques like dropout and weight penalizing are being implemented. In our work, we use singular value decomposition(SVD) for extracting the most relevant components from embeddings constructed by the network to overcome overfitting.

Authors

  • Ildar Abdrakhmanov
  • Anastasia Remizova
  • Mikhail Salnikov

About

Singular Value Decomposition for Prevention of the Graph Convolutional Network Overfitting

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published