Multichannel ALOHA Optimization for Federated Learning With Multiple Models

Large-scale wireless sensor networks are instrumental for several Internet of Things (IoT) applications involving data analytics and machine learning. The huge data volume generated by such networks imposes a change of paradigm from centralized machine learning to decentralized. Federated Learning (FL) is a well-known type of decentralized machine learning, whose efficiency heavily depends on the use of wireless communication resources. Random Access (RA) protocols, such as ALOHA, despite their simplicity, can improve the convergence time of FL systems if multiple orthogonal channels are used. This letter considers that devices are involved in the optimization of more than one model in a FL system, and then proposes an optimum method to allocate wireless resources in a multi-channel ALOHA setup. The proposed method outperforms uniform and fully-shared channel allocations in terms of convergence time.

Automated summary

This paper proposes an optimal method for allocating wireless resources in a multi-channel ALOHA setup, which outperforms uniform and fully-shared channel allocations in terms of convergence time in large-scale wireless sensor networks involving decentralized machine learning.