FedControl: When Control Theory Meets Federated Learning

TL;DR

FedControl introduces a control theory-inspired method for federated learning that adjusts client contributions based on local learning performance, improving upon traditional averaging methods like FedAvg.

Contribution

This paper presents a novel federated learning algorithm that dynamically weights client updates using control theory principles, enhancing model performance.

Findings

Outperforms FedAvg in IID settings

Adapts client contributions based on local learning performance

Extensive evaluation demonstrates improved classification accuracy

Abstract

To date, the most popular federated learning algorithms use coordinate-wise averaging of the model parameters. We depart from this approach by differentiating client contributions according to the performance of local learning and its evolution. The technique is inspired from control theory and its classification performance is evaluated extensively in IID framework and compared with FedAvg.