Chapter8. Graph Neural Networks
12기 이승현
Intro
Node Embeddings


Two key components


From "Shallow" to "Deep"

Shallow Encoders

Deep Graph Encoders


Modern ML Toolbox

Why is it Hard?
Idea : Convolutional Networks

From Images to Graphs


Real-World Graphs

A Naive Approach

Basics of Deep Learning for Graphs
Content
Setup
Graph Convolutional Networks

Idea : Aggregate Neighbors



Deep Model : Many Layers

Neighborhood Aggregation

The Math : Deep Encoder

Model Parameters

Unsupervised Training
Supervised Training


Model Design : Overview



Inductive Capability



Graph Convolutional Networks and GraphSAGE
GraphSAGE Idea

Neighborhood Aggregation


Neighbor Aggregation : Variants



Recap : GCN, GraphSAGE

Efficient Implementation

More on Graph Neural Networks

Graph Attention Networks
Simple Neighborhood Aggregation
Graph Attention Networks
Attention Mechanism


Properties of Attentional Mechanism

GAT Example : Cora Citation Net

Example Application
Application : Pinterest


Pinterest Graph

Pinsage: Overview

Embedding Nodes

Task Overview

PinSAGE Training

PinSAGE Efficiency

PinSage : Key Innovations


PinSage : Experiments

Example Pin Recommendations

PinSAGE Recommendations


General Tips and Practical Demos
General Tips

Debugging Deep Networks

Demo : Human Disease Network

Demo : Protein Interation Prediction

Last updated