liveBook · Manning

Jumping knowledge

As explained in Graph Neural Networks in Action

A graph neural network technique that combines representations from multiple neighborhood depths to adapt the effective receptive field for each node.

Jumping knowledge (JK) combines representations from multiple graph-convolution layers instead of relying only on the final layer. This lets the model use neighborhood information from different depths and adapt the effective receptive field for each node. section 3.2.2

Two-layer GCN

A CustomGCN applies two GCNConv layers, stores the activated first-layer output and second-layer output, then combines both with JumpingKnowledge(mode='cat').

How Graph Neural Networks in Action covers it

How deep it goes
Demonstrates output collection and concatenation in PyG, but does not compare alternative JumpingKnowledge modes.
Examples
Uses a two-layer CustomGCN implementation and reports baseline-versus-JK GCN classification metrics.
What it assumes
Assumes familiarity with GCN layers and node representations, while demonstrating output collection from zero.
What you can do after
Apply it via PyTorch Geometric by collecting layer outputs and passing them to a JumpingKnowledge module.
How it connects
Connects GCN layers with neighborhood aggregation and multilayer graph representations.
Cover of Graph Neural Networks in Action
liveBookmanning.com
Read Graph Neural Networks in Action on liveBook Get the book at manning.com

Related concepts

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.