Intelligent Vision-based Malware Classification using Quantised ResNets
Sashank Sridhar, Rahul Seetharaman, Sowmya Sanagavarapu · 2021 IEEE 12th Annual Information Technology, Electronics and Mobile Communication Conference (IEMCON) · 2021
Malware refers to a malicious software that exploits target system vulnerabilities for data theft, extortion or sabotage the network systems of the device. In this paper, the raw malware binary executable files are fed as images into a deep neural architecture for classification. ResNets are deep neural architectures which exhibit good feature extraction from the image data for dimensionality reduction and weight sharing, making them computationally efficient for training. Optimization of the ResNet architecture is performed using quantization of weights for building a lightweight and computationally less intensive framework for the malware classification. Quantization is achieved by converting floating-point values into integers and using a quantization function to parameterize the inputs based on the number of bits available. The deep learning framework with quantized ResNet achieved an impressive performance of 96.4% testing accuracy with 99.8% AUC score.