A graph attention network is a graph neural network architecture that aggregates information from a node’s neighbours using learned attention coefficients, allowing the model to weight each neighbour’s contribution according to its relevance. By replacing fixed or degree-normalised aggregation with attention, it adapts to local structure without requiring knowledge of the full graph in advance. Multi-head attention stabilises learning and lets the model capture several relational patterns simultaneously.
Overview
- Graph attention networks bring the attention mechanism into the message-passing framework of graph neural networks. Each node computes attention weights over its neighbours, then aggregates their features as a weighted sum, learning which connections matter for the task. This inductive, masked-attention formulation generalises to unseen graph structures and supports node- and edge-level prediction.
Mechanisms
- Attention coefficients weight neighbour contributions during aggregation
- Masked attention restricts each node to its local neighbourhood
- Multi-head attention captures multiple relational patterns and stabilises training
- Inductive formulation generalises to graphs unseen during training
- Stacked layers expand the effective receptive field over the graph
Applications
- Node classification in citation, social, and biological networks
- Link prediction and recommendation over relational data
- Molecular property prediction on graph-structured chemistry
- Knowledge-graph reasoning and entity relationship modelling