Enhancing Malware Classification with Vision Transformers: A Comparative Study with Traditional CNN Models

Abstract

Malware classification is an important task in cybersecurity, where machine learning techniques have been widely used to automate the process of identifying and categorizing malicious software. In this paper, we propose a new approach to malware classification using Vision Transformers (ViT), a state-of-the-art deep learning architecture that has shown good results in computer vision field. Our proposed ViT-based model outperforms traditional Convolutional Neural Networks (CNNs) in terms of accuracy and robustness, as it can capture long-range dependencies in the input data without relying on hand-crafted features. We evaluate our proposed model on Malimg database, and demonstrate that it achieves state-of-the-art performance, outperforming the existing traditional approaches. Furthermore, we investigate the impact of different input representations, model configurations, and training strategies on the ViT-based model's performance. Our results show that the proposed ViT-based model offers several advantages over traditional CNN models, such as better performance on large-scale and complex datasets, higher interpretability, and scalability. However, the ViT-based model requires significantly more computational resources and longer training time. Our proposed approach offers a promising direction for malware classification using ViT-based models, which can be further improved by exploring different architectures, optimization techniques, and transfer learning strategies.