D2H-AD model brings hyperdimensional computing to anomaly detection

D2H-AD, introduced by Ghazali and colleagues, is a hybrid anomaly detection framework that leverages Hyperdimensional Computing (HDC), a brain-inspired paradigm representing data as high-dimensional distributed vectors. Unlike traditional HDC methods, D2H-AD combines distance-based similarity with density-aware encoding, improving both representation and detection performance. In ablation studies, hyperdimensional encoding alone boosted ROC-AUC by up to 5.4% compared to applying the same scoring in the original feature space. The model consistently surpassed five established baselines—HDAD, ODHD, One-Class SVM, Isolation Forest, and Autoencoders—across all tested datasets, showing particular robustness to class imbalance, noise, and high complexity.

Beyond accuracy, D2H-AD is designed for resource-constrained environments. Its reliance on binary computations and a compact design results in a small memory footprint and low-latency operations, making it well-suited for TinyML and edge AI deployments. The framework is also interpretable, a key advantage for applications in healthcare, cybersecurity, smart grids, and IoT where explainability matters. This work highlights HDC's potential as an energy-efficient, scalable alternative to conventional deep learning for real-time anomaly detection on edge devices.