Downloaded 151 times
Uploaded byJungwon Kim
Graph Neural Network - Introduction
This document provides an overview of graph neural networks (GNNs). GNNs are a type of neural network that can operate on graph-structured data like molecules or social networks. GNNs learn representations of nodes by propagating information between connected nodes over many layers. They are useful when relationships between objects are important. Examples of applications include predicting drug properties from molecular graphs and program understanding by modeling code as graphs. The document explains how GNNs differ from RNNs and provides examples of GNN variations, datasets, and frameworks.
Graph Neural Network - Introduction
- 1.
- 2. Before we begin... ●It’s not about GCN (Graph Convolutional Network)! ○ Maybe explain later… ● This slide only covers basic concept of GNN
- 3. What is GraphNeural Network? ● Neural Network for Graph ○ Input: Graph ○ Output: Label ■ Number (0.95, 0.81, 0.4, …) ■ Label (Protein, Carbon-dioxide, etc.) ■ Classification (Drug/Not Drug, etc.)
- 4. Why Graph NeuralNetwork? ● Understanding of connections with many (1+) relations ○ RNN has single recursive relationship between each cell ● Representing a network ○ Self-wise Connection ○ Some objects have No Connection at all
- 5. When to useGNN? ● Chemistry ○ Drugs or not (by probability) ● Program Understanding ○ Expression Statement ○ Solve problems on programming ○ Performs better than RNN-based Models! ● Etc.
- 6.
- 7.
- 8. RNN (Recurrent NeuralNetwork) ● In RNN, each cell has cumulative information from previous cell. ○ The series can be… ■ Time Series ■ Word Order ■ Event sequence ■ Etc.
- 9. RNN (Recurrent NeuralNetwork)
- 10. GNN (Graph NeuralNetwork)
- 11. GNN (Graph NeuralNetwork)
- 12. GNN (Graph NeuralNetwork) Adjacency = 1
- 13. GNN (Graph NeuralNetwork)
- 14. GNN (Graph NeuralNetwork) ● RNN ○ Propagate by passing previous cell’s information to current cell ● GNN ○ Consider ‘adjacency’ ■ Information from cells of Adjacency = 1 ■ Information from cells of Adjacency = 2 ■ Information from cells of Adjacency = 3 ■ All the way to Adjacency = n (Maximum)
- 15. GNN Variations ● GraphNeural Network (2005) ● Spectral Networks (2014) ● Neural Message Passing (2017) ○ Gated Graph Neural Network (2016) ○ ChebyNets (2016) ○ Graph Convolution Network (2017) ● Async. Neural Message Passing (2018) ● AMPNet (2018) ● Programs As Graphs (2018) ● Etc.
- 16. Sample Datasets forGNN https://github.com/shiruipan/graph_datasets ** Not listed in link, but ‘Zachary’s karate club’ is a commonly used social network. ** (https://towardsdatascience.com/how-to-do- deep-learning-on-graphs-with-graph- convolutional-networks-7d2250723780)
- 17. Graph Nets Library(DeepMind) https://github.com/deepmind/graph_nets ● Framework from DeepMind ● Well-explained document ● Tensorflow-based
- 18. Geometric Deep Learning http://geometricdeeplearning.com/ ●Not a Framework ● Many references for GNN (Not a lot of descriptions)
- 19. References Microsoft - GraphNeural Networks: Variations and Applications (https://www.youtube.com/watch?v=cWIeTMklzNg) Graph Theory - Adjacency Matrices (https://www.ebi.ac.uk/training/online/course/network-analysis-protein-interaction-data- introduction/introduction-graph-theory/graph-0)