Adaptive Client Selection with Personalization for Communication Efficient Federated Learning

TL;DR

Experimental evaluations show that ACSP-FL minimizes the overall communication and computation overheads to train a model and converges the system efficiently, providing good convergence even in scenarios where data is distributed differently, non-independent and identical way between client devices.

Abstract

Federated Learning (FL) is a distributed approach to collaboratively training machine learning models. FL requires a high level of communication between the devices and a central server, thus imposing several challenges, including communication bottlenecks and network scalability. This article introduces ACSP-FL (https://github.com/AllanMSouza/ACSP-FL), a solution to reduce the overall communication and computation costs for training a model in FL environments. ACSP-FL employs a client selection strategy that dynamically adapts the number of devices training the model and the number of rounds required to achieve convergence. Moreover, ACSP-FL enables model personalization to improve clients performance. A use case based on human activity recognition datasets aims to show the impact and benefits of ACSP-FL when compared to state-of-the-art approaches. Experimental evaluations show that ACSP-FL minimizes the overall communication and computation overheads to train a model and converges the system efficiently. In particular, ACSP-FL reduces communication up to 95% compared to literature approaches while providing good convergence even in scenarios where data is distributed differently, non-independent and identical way between client devices.

Figures (11)

• Figure 1: Performance-based client selection with personalization: personalization phase (a) the server sends the model to the clients, the portion of the model is defined by $\mathcal{K}(\cdot)$; in evaluation phase (b) clients combine the layers shared by the server with their local model to train it with their local data and send back the performance metric to the server; selection phase (c), the server classifies clients according to performance and selects clients with lower performance than the average of received accuracies $\mathcal{A}$ to train the model in the next round of communication.
• Figure 2: Example of the decay mechanism employed by ACSP-FL.
• Figure 3: Model construction employed by ACSP-FL based on the shared and local pieces.
• Figure 4: Visualization of classes distribution by client for each dataset.
• Figure 5: Example of dynamic layer definition function based on client's accuracy.