Graph attention network
Architectures AIMech:GraphAttentionNetwork DRAFT
Aliases: GAT
A graph neural architecture that updates node representations using learned attention weights over graph neighborhoods.
Identity Notes & Concept Scope
The original GAT architecture is distinct from a generic attention layer and from graph-convolution methods that do not use its learned neighborhood attention. GATv2 and other revisions are outside this identity.
Claims & Grounded Evidence
1 Claim
Mechanism
#mechanism
Shared transformations and a learned attention function produce neighborhood scores, softmax normalizes those scores, and their weighted feature combinations form updated node representations.
⚠️ Curated Limitations & Negative Results
- The record uses the revised ICLR 2018 paper, including its specified attention nonlinearity.
- Attention coefficients are computation weights; no general causal-explanation or benchmark-superiority claim is made.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:ArtificialNeuralNetwork
Artificial Neural Network
|
broad_match | Graph attentional layers are learned neural computations, so the general artificial-neural-network target is broader. |
AIO:GraphConvolutionalNetwork
Graph Convolutional Network
|
related_match | Both use graph structure, but the GAT paper defines a specific learned attention mechanism; the mapping does not equate named GAT and GCN algorithms. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, Yoshua Bengio
|
primary paper
full_text
|
ID: arXiv:1710.10903v3
sha256:a7d811a44e1c…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T03:12:53Z
Updated: 2026-10-11T03:12:53Z
Source File:
data/curated/GraphAttentionNetwork.yaml