Deep learning in intrusion detection perspective: overview and further challenges

Abstract

Deep learning techniques are famous due to its capability to cope with large-scale data these days. They have been investigated within various of applications e.g., language, graphical modeling, speech, audio, image recognition, video, natural language and signal processing areas. In addition, extensive researches applying machine-learning methods in Intrusion Detection System (IDS) have been done in both academia and industry. However, huge data and difficulties to obtain data instances are hot challenges to machine-learning-based IDS. We show some limitations of previous IDSs which uses classic machine learners and introduce learning including feature construction, extraction and selection to overcome the challenges. We discuss some distinguished deep learning techniques and its application for IDS purposes. Future research directions using deep learning techniques for IDS purposes are briefly summarized.