A Novel Intrusion Detection System for Malware Based on Time-Series Meta-learning

In recent years, frequent occurrence of network security incidents indicates that host security is more and more fragile. However, current protection tools leads to reduce the efficiency of CPU or GPU. Meanwhile, they give up active defense and increase the security risk. Unfortunately, the existing intrusion detection systems seldom adjust the defense policy according to the host’s performance and the time when the attack might occur. Thus, different from traditional intrusion detection systems, our system is capable of intelligently detecting and predicting threats. Firstly, our system converts the malware into gray-scale images according to the instruction execution logic. Secondly, the system uses a computer vision method to identify the signature of the gray-scale images. Finally, the proposed system classifies malware family. Specifically, the system can also predict the time when a host faces a severe threat using time-series datasets and create a multi-neural network task for defending the threat. Then, a meta-learning framework is utilized to improve malware detection accuracy and defend against attacks effectively. The experimental results show that our system can accurately classify 15 malware families, and we compare our detection results with that of other IDSs, which proves that our system achieves a better performance.