Graph Neural Network

Name: Graph Neural Network
Rating: 4.4 (681 reviews)

From Graph Representation Learning to Graph Neural Network (Complete Introductory Course to GNN)

Bestseller

Created byYounes Sadat-Nejad

Last updated 6/2021

English

What you'll learn

Graph Representation Learning
Graph Neural Network (GNN)
Graph Analysis
Graph Embedding
DeepWalk
Node2Vec
Graph Convolution Network (GCN)
Graph Attention Network (GAT)
Simplifying Graph Convolution (SGC)
Inductive and Transudative Learning
GraphSAGE
Pytorch Geometric
Convolution

Course content

4 sections • 26 lectures • 4h 29m total length

Graph Definition5:24
Define graphs as nodes with features connected by edges that encode relationships; distinguish directed from undirected graphs with twitter and facebook examples, and note adjacency and weights matrices.
Storing Graph Information6:19
Learn how to store graph data for a homogeneous graph with four nodes using an edge list and an adjacency matrix, including directed versus undirected edges and weighted connections.
Graph Degree and Laplacian of Graph6:52
Explore how the degree matrix and laplacian derive from the adjacency matrix to analyze node influence and graph structure, with applications in mesh completion and brain networks.
Definition of Learning in Graph Representation Learning6:28
Embed each graph node into an embedding space to preserve node similarity. Use encoders and decoders to minimize a loss function that aligns original and embedding similarities for link prediction.
Drawback in existing graph learning models1:44
Explore the drawbacks of existing graph learning models, including lack of parameter sharing, missing semantic information, and non-inductive encoders, and learn how convolution motivates graph neural networks.
Workshop - Using Torch and Torch Geometric for defining a graph23:56
Learn to define graphs for graph neural networks using PyTorch Geometric and PyTorch, building a graph with node features, edge indices, and weights, then plot with NetworkX.

Review on Convolution Operation7:08
Graph Convolution (Signal Processing Point of View) Part A17:53
Explore how graph convolution extends signal processing concepts to graph data, using adjacency matrices and weighted shifts to apply convolution on non-Euclidean structures.
Graph Convolution (Signal Processing Point of View) Part B9:37
Explore how graph convolution uses a learnable weight matrix and shifting operator to expand from local to global graph representations across multiple layers, with k shifts and graph frequency insights.
Message Passing Framework6:00

Graph Embedding Problem Statement4:40
Learn how graph neural networks encode a graph into an embedding space via its encoder, built from aggregation and update steps, preserving similarities between original and embedded representations.
DeepWalk Algorithm8:04
Workshop - RandomWalk using karateclub library24:30
Apply random-walk based node embedding with the karateclub library on a graph to perform node classification. Generate embeddings with the Depok deep-walk model and evaluate with a logistic regression classifier.
Node2Vec Algorithm3:21
We introduce Node2Vec, a biased random walk variation of deep walk for graph embedding, using DFS and BFS to balance local and global representations.
Workshop - Node2Vec Using Karateclub10:35
Apply node2vec to the karate club graph, compare its embeddings to deepwalk, and demonstrate biased random walks with p and q to assess classification performance.
Workshop - Node2Vec Using Pytorch Geometric (Part A)15:39
Learn to compute node embeddings with node2vec in PyTorch Geometric by importing the Cora planetoid data, building a node2vec model, training with batches, and evaluating embeddings and node classification.
Workshop - Node2Vec Using Pytorch Geometric (Part B)14:35
train a node2vec model with pytorch geometric, learn lower dimensional embeddings through epochs, plot the embedding in 2d, and assess accuracy using logistic regression.
GNN Motivation4:00
Introduce simple graph convolution as the first GNN method, address the downsides of deep walks, and explain graph convolution from a signal processing perspective with the message passing framework.
Simplifying Graph Convolution Network8:56
Workshop - SGC (Part A)16:33
Workshop - SGC (Part B)16:32
Explore simplifying graph convolution networks (SGC) for node classification, configure device and optimizer, train with masks, and compare the fast SGC accuracy to traditional methods.
Graph Convolution Network (GCN)21:36
Explore graph convolution networks (gcns) using a linear projection, normalized adjacency with self-loops, and nonlinear activation to produce node embeddings across layers for semi-supervised classification.
Graph Attention Network8:36

Review on Popular GNN Embedding Methods9:04
We review the graph neural network embedding methods, compare graph convolution network and graph attention network, and explain how message passing, aggregation, updates, and learnable parameters produce node embeddings.
Transductive vs Inductive Embedding Methods2:04
GraphSAGE9:08
GraphSAGE presents inductive representation learning by aggregating neighborhood features to update node representations, enabling generalization to unseen nodes and scalable learning on large graphs.

Requirements

Introductory background on machine learning and deep learning
Introductory background on signal processing and data analysis
Algebra
Python

Description

In recent years, Graph Neural Network (GNN) has gained increasing popularity in various domains due to its great expressive power and outstanding performance. Graph structures allow us to capture data with complex structures and relationships, and GNN provides us the opportunity to study and model this complex data representation for tasks such as classification, clustering, link prediction, and robust representation.

While the first motivation of GNN's roots traces back to 1997, it was only a few years ago (around 2017), that deep learning on graphs started to attract a lot of attention.

Since the concept is relatively new, most of the knowledge is learned through conference and journal papers, and when I started learning about GNN, I had difficulty knowing where to start and what to read, as there was no course available to structure the content. Therefore, I took it upon myself to construct this course with the objective of structuring the learning materials and providing a rapid full introductory course for GNN.

This course will provide complete introductory materials for learning Graph Neural Network. By finishing this course you get a good understanding of the topic both in theory and practice.
This means you will see both math and code.

If you want to start learning about Graph Neural Network, This is for you.

If you want to be able to implement Graph Neural Network models in PyTorch Geometric, This is for you.

Who this course is for:

Engineering Graduate Students
Computer Science Graduate Students
Data Scientists
Python developers interested to learn Graph Neural Network
Deep learning engineers
Machine learning engineers
Signal Processing Engineers
Neural Network Enthusiasm

Graph Neural Network

What you'll learn

Explore related topics

Course content

Graph Terminology & Representation6 lectures • 51min

From Convolutional Neural Network to Graph Neural Network4 lectures • 41min

Introducing Different Graph Embedding Methods13 lectures • 2hr 38min

Inductive and Transductive Graph Embedding3 lectures • 20min

Requirements

Description

Who this course is for: