A Hybrid Deep Learning and Symbolic AI for Anomaly Detection in Heterogeneous High-Performance Computing Systems

Abstract

Anomaly detection in heterogeneous High Performance Computing (HPC) systems is challenging because the environment of multi-component hardware is complex. Traditional detection methods are not compatible with high-dimensional data and are heterogeneous. This study proposes a new hybrid approach by combining deep learning and symbolic artificial intelligence techniques to improve the accuracy and interpretability of anomaly detection. In this paper, the pattern recognition strengths of deep learning with the knowledge representation and reasoning of symbolic AI are combined, aiming at a unified framework that can achieve better precision in detection while realizing explainability. The framework is evaluated with HPC datasets based on performance metrics and qualitative expert assessments, bearing potential generalizability to other complex systems such as cyber-physical networks and IoT infrastructures.