At Egypt-Japan University of Science and Technology, I’m researching adversarial robustness for a Vision Transformer skin disease classifier. I evaluated One-Pixel, Two-Pixel, FGSM, and PGD attacks, and found PGD had a 100% success rate.
I built a defense pipeline using a Denoising Autoencoder and evaluated image restoration with PSNR, SSIM, MSE, GMSD, and FSIM. I also developed a DenseNet121 detector that achieved 99.04% accuracy and a 1.18% false negative rate for adversarial inputs.
My training includes network security and data science, including an AI-powered bank customer churn prediction and analysis project. I’ve also supported IT operations at Midor, assisting with network troubleshooting, cybersecurity measures, and infrastructure maintenance.

