Enhanced unknown Android Malware Detection using LG-PN: A local–global fusion approach in prototypical networks

Abstract

In malware detection research, determining whether the application has malicious intent is the most important issue. Malware variants evolve rapidly through the use of polymorphic and metamorphic techniques, posing two challenges to malware detection. First, it is very difficult to label and identify large amounts of new malware. Second, existing classification methods are usually trained on predefined malicious samples. Therefore cannot identify new types of malware. In order to solve these problems, this study proposes an innovative method based on few-shot learning, aiming to quickly adapt to new threats. This method can rely on a small number of malicious family samples to quickly infer malware that does not appear in the training set. This study conducted detection experiments on malware of unknown families, unknown samples, and unknown functions. The research results show that this method is better than existing methods when facing new malware samples.