Development of Hybrid Deep Learning Method to Detect Software Attacks
Abstract
Malicious software (malware) has been increasing. It is costly to manually detect and classify malware so that there are lots of research to automatically detect or classify malware. As malware is generated and altered to avoid detection by antivirus systems, the conventional malware detection mechanisms have suffered from several problems such as vulnerability to modified malware and poor scalability. The research aims to develop a novel deep learning method that generates data to expand a range of knowledge space for potential cyber-attacks, and detect modified and even unseen malware. With this method, we don’t have to process newly generated data in order to widen the knowledge space, and data can be directly generated and used. The method is composed of deep auto-encoders, generative adversarial networks, and transfer learning. The proposed method will be largely verified with several benchmark datasets.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 19, 2018
- Source ID
- FA23861814047
Entities
People
- Sung-Bae Cho
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- Yonsei University