Deep Learning for Security

The great success of deep learning inspires researchers to apply deep learning to security problems, for example, intrusion detection, malware detection, and software testing. However, a key challenge in applying deep learning techniques to the security domain is the lack of attack data. Unlike other fields, attack data are usually private, sensitive, and rare. Moreover, the unknown zero-day attacks even do not have existing data or features to construct a model. To this end, we proposed to research the feasibility of deep learning in practical security scenarios where no attack data are needed, i.e., anomaly detection. We show that a deep learning model can be trained to predict the normal behavior of a system and investigate the distribution of the model prediction errors. Specifically, we investigate anomaly detection in multiple domains, e.g., power-grid infrastructures, smartphone impostor detection, and cloud anomaly detection.


Deep learning for anomaly detection

This project aims to explore deep learning techniques for anomaly detection.
  1. Zecheng He, Aswin Raghavan, Guangyuan Hu, Sek Chai and Ruby B. Lee "Power-grid controller anomaly detection with enhanced temporal deep learning", IEEE International Conference On Trust, Security And Privacy In Computing And Communications (TrustCom), 2019
  2. Guangyuan Hu, Zecheng He, and Ruby B. Lee, "Smartphone Impostor Detection with Behavioral Data Privacy and Minimalist Hardware Support", TinyML Research Symposium, 2021