FoggyTrust Enhances Federated Learning Robustness with Hierarchical Trust

Federated Learning (FL) aims to train models collaboratively across distributed clients while protecting private data, but it's vulnerable to malicious or corrupted clients (Byzantine attacks). Existing solutions like FLTrust use a server-side root dataset to assign trust scores to client updates, but struggle with globally heterogeneous data distributions. FOGGYTRUST is proposed as a hierarchical extension to FLTrust, designed to enhance robustness. It localizes trust computation to "fog nodes," which allows the framework to better manage data heterogeneity by grouping clients into locally homogeneous clusters. This two-level architecture addresses both distribution mismatch in trust estimation and client drift across groups. By combining local trust-based aggregation with global optimizers like FedAdam and SCAFFOLD, FOGGYTRUST achieves significant gains, especially in challenging heterogeneous settings. On CIFAR-10 under Krum and Trim attacks, it showed over 50% improvement compared to FLTrust. Its promise extends to safety-critical applications like distributed wildlife monitoring, demonstrating the value of hierarchical trust networks in robust federated learning.