Track: Research Track |
Towards Trustworthy Neural Network Intrusion Detection for Web SQL Injection |
Due to the pervasion of Artificial Intelligence, neural network has been widely used in various domains. However, trust issues have rise for the decision made by neural network models, because they are opaque and cannot explain their decisions. Numerous explainable artificial intelligence methods have been proposed to solve this question, but most of them can provide vague explanations. In security-centered domains such as cybersecurity, which relay on binary string analysis, even one bit’s misinterpretion will cause tremendous misunderstanding and misleading. Thus, formal and rigorous explanations are imperative. This paper proposes an rigorous explainable Web SQL Injection intrusion detection based on neural network models. Prime Implicant explanations that are 100% loyal to the model are extracted. Explanation performance are presented and compared with current explainable AI methodology SHAP in terms of precision and time overhead in detail. It is evident that proposed explainable neural network model are tractable and scalable. |
|
Presentation Video |
Presentation Notes |
QianruZhou-trustworthyAI.pdf |