Graph attention network v2
Architectures AIMech:GATv2 DRAFT
Aliases: GATv2
A graph neural architecture whose neighborhood attention scorer applies a learned scalar projection after a nonlinear transformation of the query-neighbor pair, allowing attention rankings to depend on the query node.
Identity Notes & Concept Scope
GATv2 changes the scoring-function operation order of the original GAT. It is a distinct architecture, not a software version label or a synonym for every graph attention network.
Claims & Grounded Evidence
1 Claim
Mechanism
#architecture-mechanism
Concatenate query and neighbor features, apply a learned affine transformation and LeakyReLU, and then project to an attention score; normalize scores over the neighborhood with softmax and use them to aggregate transformed neighbor features.
⚠️ Curated Limitations & Negative Results
- Dynamic attention is an expressiveness property of the parameterized scorer, not a guarantee that any trained instance learns useful node-specific rankings.
- This record supports the architecture and operation order, not universal benchmark superiority or a causal interpretation of attention coefficients.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:ArtificialNeuralNetwork
Artificial Neural Network
|
broad_match | The specified graph attentional layers form an artificial neural architecture, so the artificial-neural-network target is broader. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Shaked Brody, Uri Alon, Eran Yahav
|
primary paper
full_text
|
ID: arXiv:2105.14491v3
sha256:d316a7d23d73…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T04:11:44Z
Updated: 2026-10-11T04:11:44Z
Source File:
data/curated/GATv2.yaml