Task-Aware Meta Learning-Based Siamese Neural Network for Classifying Control Flow Obfuscated Malware

Malware authors apply different techniques of control flow obfuscation, in order to create new malware variants avoid detection. Existing Siamese neural network (SNN)-based detection methods fail correctly classify families when such obfuscated samples are present the training dataset, resulting high false-positive rates. To address this issue, we propose a novel task-aware few-shot-learning-based Neural Network that is resilient against presence affected by obfuscation techniques. Using average entropy features each family as inputs, addition image features, our model generates parameters for feature layers, more accurately adjust embedding families, which has variants. In addition, proposed method can classes, even if there only one or few available. Our utilizes few-shot learning with extracted pre-trained (e.g., VGG-16), bias typically associated trained limited number samples. approach highly effective recognizing unique signatures, thus classifying belong same family, experimental results, validated N-way on N-shot learning, show classification accuracy, exceeding rate \textgreater 91\%, compared other similar methods.