Synthetic Data Generation Without Real Data: Uncovering Insights in Malware Detection

The use of synthetic data for training machine learning (ML) models has garnered significant attention among researchers as a potential solution to the challenge of balancing privacy protection and data utilization. This paper introduces a novel approach for generating synthetic data that specifically addresses this challenge. Unlike existing methods that focus on closely replicating real data distributions, our proposed approach aims to generate synthetic data without directly using real data, while still enabling the training of ML models to extract specific bits of information. This can be achieved by leveraging only general knowledge about the problem domain, acquired without accessing real data. We applied this approach to the task of malware detection and conducted experiments to evaluate its effectiveness. The results not only validated the efficacy of our proposed approach but also led to a significant discovery in the field of malware detection.