International Journal of Scientific Research and Engineering Development

This study addresses vulnerabilities in digital memory forensics caused by adversarial attacks such as page table corruption, frame modification, and malicious process injection, which often reduce reconstruction quality and compromise forensic reliability. The research aims to develop a robust and adversarially resilient Generative Adversarial Network (GAN)-based framework for accurate memory reconstruction while maintaining structural consistency under hostile conditions. The proposed Robust GAN integrates adversarial training, memory-aware perturbation modeling, and structural consistency constraints to improve reconstruction performance. Evaluation was conducted using the DFRWS Challenge Dataset and synthetic KVM memory dumps containing about 1,250,000 memory frames under both clean and adversarial conditions. Attack scenarios included page table noise, frame corruption, and malicious process injection. Performance was measured using Reconstruction Accuracy, Robustness Score, Peak Signal-to-Noise Ratio (PSNR), Mean Squared Error (MSE), Structural Consistency Score (SCS), and Attack Sensitivity Index (ASI). Results show that the Robust GAN outperformed the baseline GAN, achieving 70% reconstruction accuracy compared to 50%, while also demonstrating improved robustness, lower reconstruction errors, and stronger preservation of memory structural integrity.

Onyagu Chika Lilian, Snow Ngozi Monye, Izunna Lucky Chibuike, Usih Mary Emetena, Egheneji Akpevwe Wealth,"Adversarial Robustness in AI-Driven Hypervisor Memory Reconstruction" International Journal of Scientific Research and Engineering Development, V9(3): Page(884-892) May-June 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.