# SAVAGE
**Entity class:** Adversarial graph-learning framework
**Aliases:** Sparse Vicious Attacks on Graph Networks
**Maturity:** Developed
## Definition
**SAVAGE** demonstrates sparse adversarial attacks on graph neural networks by introducing carefully selected malicious nodes or edges. It shows how small graph changes can manipulate link-prediction output, a direct warning against treating predicted associations as evidence.
## Relationships
- **attacks:** [[wiki/Graph Neural Network|Graph Neural Network]] and [[wiki/Link Prediction|Link Prediction]].
- **risk to:** [[wiki/Counterterrorism Intelligence|Counterterrorism Intelligence]] and [[wiki/Entity Resolution|Entity Resolution]].
- **cybersecurity relevance:** [[wiki/Cybersecurity|Cybersecurity]].
## Sources / Provenance
- [IEEE — Sparse Vicious Attacks on Graph Neural Networks, 2024](https://www.computer.org/csdl/journal/ai/2024/05/10264111/1QKPb4aifoA)