As malware continues to evolve, deep learning models are increasingly used for malware detection and classification, including image-based classification. However, adversarial attacks can be used to perturb images so as to evade detection by these models. This study investigates the effectiveness of training deep learning models with Generative Adversarial Network-generated data to improve their robustness against such attacks. Two image conversion methods, byteplot and space-filling curves, were used to represent the malware samples, and a ResNet-50 architecture was used to train models on the image datasets. The models were then tested against a projected gradient descent attack. It was found that without GAN-generated data, the models’ prediction performance drastically decreased from 93-95% to 4.5% accuracy. However, the addition of adversarial images to the training data almost doubled the accuracy of the models. This study highlights the potential benefits of incorporating GAN-generated data in the training of deep learning models to improve their robustness against adversarial attacks.
25 June 2025 in European Conference on Cyber Warfare and Security
The increasing complexity and sophistication of malware pose significant challenges to traditional detection techniques. Conventional methods like signature-based detection are ineffective against advanced threats such as polymorphic and zero-day malware. This research investigates the application of Dynamic Graph Neural Networks (DGNNs) for malware detection using a dataset of API call sequences. DGNNs, an advanced form of Graph Neural Networks, are capable of modeling dynamic graphs, capturing both the temporal and structural evolution of API interactions. Using these strengths, the study develops and evaluates a DGNN-based framework designed to effectively distinguish between benign and malicious behavior in real time, demonstrating its suitability for detecting complex, evolving malware patterns. The results show that DGNN outperform traditional machine learning models in detecting complex malware patterns, achieving high accuracy of up to 97%, F1 scores of up to 98% in unbalanced datasets, and competitive results in balanced datasets. The models also achieved ROC-AUC scores exceeding 97% in specific configurations, highlighting their effectiveness in identifying advanced malware pat- terns and resilience against novel threats. Although challenges in scalability and computational complexity remain, this work proposes potential solutions to enhance practical implementation. These findings highlight the potential of DGNNs to transform malware detection and significantly improve endpoint security, making them a promising tool for addressing the evolving challenges of modern cybersecurity.