Evidential Trust-Aware Model Personalization in Decentralized Federated Learning for Wearable IoT

TL;DR

Murmura addresses personalized learning in decentralized federated learning for wearable IoT under non-IID data. It introduces evidential deep learning to quantify epistemic uncertainty and derive peer compatibility scores for trust-aware, selective aggregation in a fully decentralized setup. The approach enables fast convergence and robust personalization by excluding incompatible peer updates while leveraging compatible ones, demonstrated on three wearable datasets with strong performance gains and stability across hyperparameters. The open-source Murmura framework provides a modular platform for evaluating decentralized personalization under controlled heterogeneity and varied network topologies.

Abstract

Decentralized federated learning (DFL) enables collaborative model training across edge devices without centralized coordination, offering resilience against single points of failure. However, statistical heterogeneity arising from non-identically distributed local data creates a fundamental challenge: nodes must learn personalized models adapted to their local distributions while selectively collaborating with compatible peers. Existing approaches either enforce a single global model that fits no one well, or rely on heuristic peer selection mechanisms that cannot distinguish between peers with genuinely incompatible data distributions and those with valuable complementary knowledge. We present Murmura, a framework that leverages evidential deep learning to enable trust-aware model personalization in DFL. Our key insight is that epistemic uncertainty from Dirichlet-based evidential models directly indicates peer compatibility: high epistemic uncertainty when a peer's model evaluates local data reveals distributional mismatch, enabling nodes to exclude incompatible influence while maintaining personalized models through selective collaboration. Murmura introduces a trust-aware aggregation mechanism that computes peer compatibility scores through cross-evaluation on local validation samples and personalizes model aggregation based on evidential trust with adaptive thresholds. Evaluation on three wearable IoT datasets (UCI HAR, PAMAP2, PPG-DaLiA) demonstrates that Murmura reduces performance degradation from IID to non-IID conditions compared to baseline (0.9% vs. 19.3%), achieves 7.4$\times$ faster convergence, and maintains stable accuracy across hyperparameter choices. These results establish evidential uncertainty as a principled foundation for compatibility-aware personalization in decentralized heterogeneous environments.

Figures (5)

• Figure 1: Murmura framework architecture showing the three-layer design. The Configuration Layer provides YAML/JSON-based experiment specification. The Core Components layer contains the Network Orchestrator and Node (Local Learner), along with Topology and Model definitions. The Pluggable Modules layer enables flexible integration of aggregation strategies and data partitioning schemes.
• Figure 2: Model accuracy across data heterogeneity levels (Dirichlet $\alpha$), averaged across all three datasets. Lower $\alpha$ indicates higher heterogeneity. Murmura (Evidential Trust) maintains consistent performance as heterogeneity increases.
• Figure 3: Performance degradation from IID ($\alpha=1.0$) to non-IID ($\alpha=0.1$) conditions. Lower values indicate better robustness to heterogeneity. Murmura shows minimal degradation (0.9%) compared to baselines.
• Figure 4: Model personalization under high heterogeneity ($\alpha=0.1$). Bars show mean accuracy; error bars show standard deviation across nodes. Murmura achieves high accuracy with low variance across nodes.
• Figure 5: Convergence speed comparison showing rounds to reach peak accuracy. Murmura converges 7.4$\times$ faster than FedAvg by filtering incompatible peer updates.