Core Concepts
SHERD method enhances robustness and performance in graph neural networks by identifying vulnerable nodes.
Abstract
The paper introduces the SHERD method, which focuses on enhancing the robustness and performance of graph neural networks. By leveraging early training representations, SHERD identifies vulnerable nodes susceptible to adversarial attacks. The method aims to create a robust subgraph that maintains node classification accuracy while eliminating vulnerable nodes. Experiments conducted across various datasets demonstrate the effectiveness of SHERD in improving model accuracy and resilience against adversarial attacks. The approach is extensible, efficient, and outperforms standard compression techniques in terms of accuracy and robustness.
Stats
Nodes: 2708
Features: 1433
Edges: 10556
Quotes
"Our experiments demonstrate the increased performance of SHERD in enhancing robustness by comparing the network’s performance on original and subgraph inputs against various baselines alongside existing adversarial attacks."
"To address vulnerabilities and overcome limitations associated with incomplete graph information, it is imperative to identify and eliminate susceptible nodes."