Enhancing the Resilience of Privacy-Preserving Machine Learning using Adversarial Techniques

This paper introduces a novel approach to enhance privacy-preserving machine learning (PPML) by integrating adversarial techniques with Homomorphic Encryption (HE) and Differential Privacy (DP). Privacy-Preserving machine learning (PPML) plays a key role in privacy protection. Current methods like homomorphic encryption (HE) and differential privacy (DP) aim to strike a balance between keeping data private and making sure models work well. This method embeds adversarial attacks within model optimization, strengthening resistance to privacy breaches while maintaining high model performance. Experimental results across various datasets achieving an accuracy of 89% in HE and DP demonstrate that this approach effectively balances privacy and utility, outperforming traditional PPML methods in safeguarding sensitive data without compromising model accuracy.