Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• PEANUT is a simple, gradient-free, restricted black-box attack that injects virtual nodes to capitalize on GNN vulnerabilities.method 95%
• PEANUT is an injection-based attack that is considered more practical and realistic than graph modification attacks.method 95%
• PEANUT operates at the inference phase as an evasion attack and is applicable almost immediately without lengthy optimizations or parameter learning.method 95%
• GNN performance can be significantly deteriorated even with injected nodes that have zero features, highlighting the importance of connectivity design.result 95%
• The proposed attack, PEANUT, is effective on real-world datasets across three graph tasks despite its simplicity.result 95%
• The vulnerability exploited by PEANUT stems from GNN architectures that explicitly use graph topology matrices (adjacency or Laplacian) for message passing.technical 95%
• The perturbation S*v which maximizes L(S v) while satisfying S*v S*v^T F <= Delta is given by S*v = Delta * u1 * v, where u1 is the dominant eigenvector of HH^T and v is any unit-norm vector.technical 95%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card