Utility Fairness for the Differentially Private Federated Learning

TL;DR

This paper addresses utility unfairness in federated learning over IoT networks by proposing a differential privacy-based control scheme that balances device contribution and expenditure, significantly reducing energy cost disparities.

Contribution

It introduces a novel differential privacy approach combined with adaptive policies to mitigate utility unfairness caused by device heterogeneity in federated learning.

Findings

Reduces energy cost standard deviation by 99% among devices

Balances training loss variation around 0.103

Addresses device heterogeneity and strategic behavior in FL

Abstract

Federated learning (FL) allows predictive model training on the sensed data in a wireless Internet of things (IoT) network evading data collection cost in terms of energy, time, and privacy. In this paper, for a FL setting, we model the learning gain achieved by an IoT device against its participation cost as its utility. The local model quality and the associated cost differs from device to device due to the device-heterogeneity which could be time-varying. We identify that this results in utility unfairness because the same global model is shared among the devices. In the vanilla FL setting, the master is unaware of devices' local model computation and transmission costs, thus it is unable to address the utility unfairness problem. In addition, a device may exploit this lack of knowledge at the master to intentionally reduce its expenditure and thereby boost its utility. We propose to control the quality of the global model shared with the devices, in each round, based on their contribution and expenditure. This is achieved by employing differential privacy to curtail global model divulgence based on the learning contribution. Furthermore, we devise adaptive computation and transmission policies for each device to control its expenditure in order to mitigate utility unfairness. Our results show that the proposed scheme reduces the standard deviation of the energy cost of devices by 99% in comparison to the benchmark scheme, while the standard deviation of the training loss of devices varies around 0.103.