Fisher-PSO Latent Optimizer: Metaheuristic-Based Optimization for Privacy-Utility Tradeoff in WGAN

Recently, the challenge of generating high-quality synthetic data and ensuring privacy has gained significant attention in sensitive domains such as healthcare. This paper introduces Fisher-PSO Latent Optimizer, an advanced model which incorporates differential privacy and Niche Particle Swarm Optimization (PSO) into Wasserstein Generative Adversarial Network (WGAN). The primary objective is to optimize the latent vectors to generate synthetic data with high fidelity while effectively balancing the trade-off between privacy and utility. Stochastic gradient descent (SGD) maintains differential privacy through controlled noise addition and ensures strict privacy guarantees while preserving essential data characteristics. Experimental results show that the Fisher PSO latent optimizer demonstrated better performance than conventional Differentially Private Generative Adversarial Network (DP-GAN) model in terms of privacy-utility trade-offs, especially under strict privacy budget, as well as higher Area Under the Curve - Receiver Operating Characteristic (AUROC) and stability.