Design and Implementation of Adaptive Personalized Differential Privacy in Federated Learning

Federated Learning enables collaborative model training across distributed clients while keeping raw data local, making it attractive for privacy sensitive domains such as healthcare, finance, and edge intelligence. This paper proposes the Design and Implementation of Adaptive Personalized Differential Privacy framework for federated learning that dynamically allocates client specific privacy budgets based on quantified contribution scores. The proposed approach is implemented using PyTorch and Opacus and evaluated on benchmark datasets under non independent and identically distributed data settings. Experimental results demonstrate that the adaptive personalised strategy achieves higher global model accuracy, improved fairness across clients, and a more efficient privacy utility balance compared to uniform differential privacy baselines. Overall, the study confirms that adaptive personalised differential privacy significantly enhances the practicality and robustness of federated learning systems.