Graph machine learning

As explained in Graph Neural Networks in Action

Graph-based learning applies machine learning to data represented as a graph and learns from its structural relationships.

Graph machine learning applies machine learning to graphs: structured data whose nodes, edges, and features encode relationships. It builds representations that support node prediction, link prediction, graph classification, visualization, clustering, and other analysis tasks. section 1.2 section 1.2.3

A practical graph-learning workflow typically follows these steps:

• Represent entities as nodes, relationships as edges, and available attributes as graph features. section 1.2 section 8.0
• Convert raw or nongraph data into edge lists, adjacency lists, and model-ready graph data. section 8.0
• Create node, edge, or whole-graph embeddings as lower-dimensional representations that preserve useful patterns and relationships. section 1.2.3 section 2.0
• Train a suitable method for the target task, including classical algorithms, matrix methods, or graph neural networks. section 1.2.3 section 2.0

Graph neural networks are one graph-learning approach. They combine neural networks with graph data, transform nodes or edges into nonlinear embeddings through message passing, and learn representations during training. Their permutation invariance makes predictions independent of node ordering or adjacency-matrix arrangement, while their design is less suited to images and tables. section 1.2.4

Recruiting-firm social graph

A recruiting firm can model job candidates as nodes and relationships between candidates as edges. Candidate IDs, job types, and industries become node data; raw files are transformed into graph structures and prepared with NetworkX and PyTorch Geometric for GNN training. section 8.0

Figure 1.7 An overview of how GNNs work. An input graph is passed to a GNN. The GNN then uses neural networks to transform graph features such as nodes or edges into nonlinear embeddings through a process known as message passing. These embeddings are then tuned to specific unknown properties using training data. After the GNN is trained, it can predict unknown features of a graph.

Follow the pipeline from an input graph through message passing and learned embeddings to predictions of unknown graph features. section 1.2.4

How Graph Neural Networks in Action covers it

How deep it goes

Defines graph-learning families, embeddings, GNN message passing, and preprocessing workflows without deriving their mathematics.

Examples

Uses Titanic-derived data, social networks, biological networks, and a recruiting-firm candidate graph.

What it assumes

Assumes basic machine-learning familiarity while introducing essential graph concepts, terminology, use cases, and GNN mechanics.

What you can do after

Apply graph workflows with NetworkX and PyTorch Geometric, from raw data conversion through graph preparation and GNN training.

How it connects

Connects graphs with embeddings, node prediction, link prediction, graph classification, message passing, and graph neural networks.

liveBookmanning.com

Read Graph Neural Networks in Action on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Optimization Algorithms: AI techniques for design, planning, and control problemsGraph machine learning is machine learning on graph-structured data that models relationships among entities as well as entity attributes.
• 

  liveBookmanning.com
  Graph Algorithms for Data ScienceGraph machine learning is machine learning that represents data and their relationships as a graph and exploits graph structure during modeling.

Compare how 3 books explain graph machine learning →

Related concepts

• machine learningkind of · Machine learning is a field of computational methods in which systems learn patterns or relationships from data to make predictions, decisions, or other outputs.
• graph neural networkused by · A graph neural network is a neural-network model that learns representations or predictions from graph-structured data by combining information from connected entities.
• embeddingproduces · A numerical vector representation of an item or input that encodes features or relationships useful for computational tasks.
• deep learningkind of · Deep learning is a machine-learning approach that uses neural networks with multiple computational layers to learn representations from data.
• permutation invarianceuses · Permutation invariance is the property that a computation’s result remains unchanged when equivalent input elements are reordered.
• local neighborhooduses · A local neighborhood is a set of elements considered near a specified element according to a defined spatial, topological, distance-based, or relational criterion.

This explainer was generated by Manning's knowledge graph from the text of Graph Neural Networks in Action. The section links point to the parts of the book it draws on.