Though Graph Neural Networks (GNNs) are relatively newer deep learning methods, they have gained immense popularity in the research space in the last few years. GNNs can be applied to graphs to conduct node-level, graph-level and edge-level predictions.
Let us have a look at a few of the courses and tools to get you started with understanding more about GNNs.
CS224W: Machine Learning with Graphs – Stanford University
The course will teach students topics such as representation learning and graph neural networks, algorithms for the World Wide Web, and reasoning over Knowledge Graphs. It will also cover areas related to influence maximisation, disease outbreak detection, and social network analysis. Students will be introduced to machine learning techniques and data mining tools apt to reveal insights on a variety of networks.
In order to pursue this course, students are expected to have the following background.
- Knowledge of basic computer science principles sufficient to write a reasonably non-trivial computer program
- Familiarity with the basic probability theory
- Familiarity with the basic linear algebra
For more information, click here.
Graph Neural Network: Udemy
It provides a full introductory course for GNN. It will cover topics such as 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 and Convolution.
The course is suitable for engineering graduate students, computer science graduate students, data scientists, Python developers interested in learning Graph Neural Network, deep learning engineers, machine learning engineers, signal processing engineers and neural network enthusiasts.
The students should have an introductory background in machine learning and deep learning, introductory background in signal processing and data analysis, algebra and Python.
For more information, click here.
GNNs: An introduction to Graph Neural Networks – Skillsoft
This course will teach students various use cases for machine learning in analysing graph data and discuss the challenges around modelling graphs for use in neural networks. It will show how a convolution function captures the properties of a node and those of its neighbours. The students will also get to work with the Spektral Python library to model a graph dataset for application in a GNN.
They will also practice defining a convolution function for a GNN and examining how the resultant message propagation works. The course will help participants understand the need for node embeddings in setting up graphs for machine learning and talk about how neural networks are constructed and applied to graph data.
For more information, click here.
Libraries
Deep Graph Library (DGL)
With the Deep Graph Library, one can build models with PyTorch, TensorFlow or Apache MXNet. It provides memory-efficient message-passing primitives for training Graph Neural Networks. One can scale to giant graphs via multi-GPU acceleration and distributed training infrastructure.
Graph Nets
It is DeepMind’s library for building graph networks in TensorFlow and Sonnet. The installation is compatible with Linux/Mac OS X and Python 2.7 and 3.4+. It works for both the CPU and GPU versions of TensorFlow. However, it does not list TensorFlow as a requirement; the user has to install TensorFlow separately for the library to work.
PyTorch Geometric
PyG (PyTorch Geometric) is a library built on PyTorch to write and train GNNs. The library has easy-to-use mini-batch loaders for operating on many small and single giant graphs, multi GPU-support, distributed graph learning through Quiver and a large number of common benchmark datasets.
