Secure Data Augmentation in Deep Learning for Enhanced Network Protection
Khushbu Kriplani, Anil Pratap Singh, Chetan Shingadiya, Naim Ansari · 2023
The proposed methodology, titled “Secure Data Augmentation for Enhanced Network Protection,” presents a comprehensive approach to augmenting training data in a secure and privacy-preserving manner while strengthening network protection. This approach revolves around three key algorithms: Secure Data Transformation, Privacy-Preserving Data Generation, and Adversarial Training for Robustness. The first algorithm, Secure Data Transformation, focuses on augmenting the original training dataset while ensuring privacy preservation through a privacy-preserving transformation function. This controlled noise addition diversifies the data, crucial for robust deep learning models. The second algorithm, Privacy-Preserving Data Generation, leverages homomorphic encryption to augment encrypted training data securely. The third algorithm, Adversarial Training for Robustness, enhances model resilience against adversarial attacks by incorporating adversarial examples during training. The proposed methodology has shown superior performance across various evaluation metrics, showcasing its potential for bolstering network protection.