Characterize interactions between IoC encoding and federated regularization

Characterize how the FedIoC IoC-contrastive gradient-encoding component interacts with proximal regularization in FedProx and variance-reduction control variates in SCAFFOLD, including the conditions that determine whether campaign recovery and classification improve or deteriorate.

Background

FedIoC is designed as a modular client-side gradient encoder that can be combined with different federated-learning aggregation methods. The paper evaluates combinations with FedProx and SCAFFOLD, which respectively constrain local model drift and reduce client variance through control variates.

The empirical effects are inconsistent: the contrastive component improves some FedProx campaign-recovery results but interacts poorly with the already unstable SCAFFOLD configuration. The authors explicitly leave open a systematic characterization of these interactions.

References

In our experiments the contrastive term composes acceptably with proximal regularization (FedProx) but interacts poorly with variance reduction (SCAFFOLD), which is unstable in our regime; characterizing these interactions is left open.

— Federated Attack Campaign Detection via Contrastive Encoding of Threat Indicators in Gradient Updates  (2609.04815 - Röder et al., 4 Sep 2026) in Section 7, “FL Aggregation under Heterogeneity”