Federated Learning is a distributed learning paradigm that faces two key challenges: systems heterogeneity (variations in device capabilities) and statistical heterogeneity (non-identical data distributions). This paper introduces FedProx, a framework that addresses these challenges by generalizing and re-parameterizing FedAvg, the current state-of-the-art method for federated learning. FedProx allows devices to perform variable amounts of local work based on their systems constraints and incorporates a proximal term to improve stability. Theoretically, the paper provides convergence guarantees for FedProx in non-identical data distributions while considering device-level systems constraints. Empirically, FedProx demonstrates improved convergence and accuracy compared to FedAvg, especially in highly heterogeneous settings, with an average improvement of 22% in test accuracy. The paper also discusses the choice of the penalty constant \(\mu\) in the proximal term and its adaptive tuning based on model performance.Federated Learning is a distributed learning paradigm that faces two key challenges: systems heterogeneity (variations in device capabilities) and statistical heterogeneity (non-identical data distributions). This paper introduces FedProx, a framework that addresses these challenges by generalizing and re-parameterizing FedAvg, the current state-of-the-art method for federated learning. FedProx allows devices to perform variable amounts of local work based on their systems constraints and incorporates a proximal term to improve stability. Theoretically, the paper provides convergence guarantees for FedProx in non-identical data distributions while considering device-level systems constraints. Empirically, FedProx demonstrates improved convergence and accuracy compared to FedAvg, especially in highly heterogeneous settings, with an average improvement of 22% in test accuracy. The paper also discusses the choice of the penalty constant \(\mu\) in the proximal term and its adaptive tuning based on model performance.