A Deep Learning-Based Model for Malware Detection Using Transformer Architecture

This research presents a new Transformer-based deep learning malware detection system for Portable Executables on Windows. The system analyzed a dataset of 11,000 executables, containing 8,700 malware and 2,300 benign files from VirusTotal, and demonstrated the ability of Transformers to learn highly complex patterns from assembly code. The system's multilevel architecture includes disassembly, tokenization, and feature extraction, allowing it to discriminate between malicious and benign files with great accuracy. The Transformer model, through its self-attention mechanisms, achieves better learning dynamics and higher accuracy compared to traditional models. Preliminary results show high precision in detecting malware, indicating the potential for Transformers to significantly enhance cybersecurity and address increasing threats.