Edge AI in Highly Volatile Environments: Is Fairness Worth the Accuracy Trade-off?

TL;DR

This paper tackles the problem of balancing model accuracy and participation fairness in federated learning deployed over highly volatile edge environments. It evaluates fairness-based client selection strategies RBFF and RBCSF against random and greedy baselines across CIFAR10, FashionMNIST, and EMNIST, under IID and non-IID data distributions and static as well as dynamic resource scenarios. The findings indicate that fairness-aware approaches can improve equitable client involvement with only modest impacts on accuracy and convergence speed, while naive greedy methods often boost speed at the cost of fairness and performance; volatility can further amplify the benefits of fairness mechanisms. The work provides practical insights for designing robust, fair FL at the edge and suggests directions for integrating privacy guarantees and extending fairness beyond per-round participation to longer-term, cross-round considerations.

Abstract

Federated learning (FL) has emerged as a transformative paradigm for edge intelligence, enabling collaborative model training while preserving data privacy across distributed personal devices. However, the inherent volatility of edge environments, characterized by dynamic resource availability and heterogeneous client capabilities, poses significant challenges for achieving high accuracy and fairness in client participation. This paper investigates the fundamental trade-off between model accuracy and fairness in highly volatile edge environments. This paper provides an extensive empirical evaluation of fairness-based client selection algorithms such as RBFF and RBCSF against random and greedy client selection regarding fairness, model performance, and time, in three benchmarking datasets (CIFAR10, FashionMNIST, and EMNIST). This work aims to shed light on the fairness-performance and fairness-speed trade-offs in a volatile edge environment and explore potential future research opportunities to address existing pitfalls in \textit{fair client selection} strategies in FL. Our results indicate that more equitable client selection algorithms, while providing a marginally better opportunity among clients, can result in slower global training in volatile environments\footnote{The code for our experiments can be found at https://github.com/obaidullahzaland/FairFL_FLTA.

Figures (6)

• Figure 1: Impact of number of clients on accuracy JFI tradeoff under volatile resources
• Figure 2: Impact of percentage of selected clients on accuracy JFI tradeoff under volatile resources
• Figure 3: Performance comparison in highly volatile edge environments with statistical heterogeneity (Class Non-IID, 50 clients, dynamic). This scenario represents the intersection of edge intelligence constraints, high volatility, and maximum fairness challenges. RBFF demonstrates superior fairness (JFI = 0.99X) while maintaining competitive accuracy in the most challenging experimental condition.
• Figure 4: Baseline performance in stable edge environments for volatility impact assessment (Class Non-IID, 50 clients, static). Direct comparison with Figure \ref{['fig:class_noniid_50_dynamic']} enables precise quantification of volatility effects on fairness-aware client selection algorithms.
• Figure 5: Fairness algorithm performance under uniform data distribution in volatile edge environments (IID, 50 clients, dynamic). This scenario illustrates optimal conditions for fairness mechanisms, demonstrating how algorithms adapt when statistical heterogeneity is minimized while environmental volatility persists.