AI-DRIVEN TECHNIQUES FOR MALWARE AND MALICIOUS CODE DETECTION

ABSTRACT

Driven by considerable economic profits, there has been explosive growth of malware which has posed significant damages and finical losses to public users. The increasingly sophisticated malware call for new defensive techniques that are capable of protecting legitimate users against novel threats in cyberspace. To solve this problem, in this dissertation, we aim to design and develop innovative links between artificial intelligence (AI) and cybersecurity for new defensive techniques against evolving malware attacks and thus help improve security in cyberspace. More specifically, we first study how deep learning models integrating cybersecurity domain knowledge can be designed for intelligent malware detection. In order to comprehensively characterize the complex semantics and social relations among various applications (apps) for malware detection, we propose to utilize both content- and relation-based features and construct structured heterogeneous information networks (i.e., heterogeneous graphs) for data abstraction. Based on the constructed graphs, we propose novel node representation learning techniques and further address the out-of-sample node representation learning problem in malware detection. As the representations of malware could be highly entangled with benign apps in the complex ecosystem of development, to solve this problem, we propose an innovative adversarial disentangler to separate the distinct, informative factors of variations in the observed environment of given apps for evolving malware detection. In addition, we bring an important new insight to exploit social coding properties for the study of code security problem on social coding platforms such as Stack Overflow and GitHub. Last but not least, as AI-driven systems have been widely used in various applications including automatic malware detection, their success may also incentivize attackers to defeat the learning-based models to bypass the detection. To address this challenge, we x 1 further investigate adversarial learning over heterogeneous graph based models to improve their robustness in malware detection. To comprehensively evaluate the proposed techniques for malware and malicious code detection, we conduct extensive experimental studies based on the real-world data collections. The promising results demonstrate the outstanding performance of our proposed models compared with state-of-the-art approaches. Our proposed techniques and developed models can be readily applied to realworld malware and malicious code detection tasks.

BUY THIS TOPIC ON WHATSAPP

Related Project Materials

Background to the Study

Mathematics and its applications have evolved significantly. The introduction of computation and...

ABSTRACT

In several quarters, there have been allegation making round, that state governments u...