Federated Learning (FL) allows clients to collaboratively train models without centralizing data, making it a widely adopted privacy-enhancing technology (PET). However, FL remains vulnerable to orchestrator-driven privacy attacks.
This paper studies an underexplored threat where a dishonest orchestrator intentionally manipulates the aggregation process to induce targeted overfitting in specific clients' local models. While prior work focuses on reducing information leakage during training, we emphasize the importance of early client-side detection of targeted overfitting, enabling clients to disengage before significant harm occurs.
To this end, we propose three detection techniques — label flipping, backdoor trigger injection, and model fingerprinting — which enable clients to verify the integrity of the global aggregation.
We evaluated our methods across multiple datasets and attack scenarios. In single-client attacks, all three methods detect orchestrator-induced overfitting within 1-2 training rounds with F1 scores up to 0.7. Scalability experiments further show that detection effectiveness is influenced by cohort composition and method parameters.
These results demonstrate that client-side integrity testing can provide early, effective, and scalable detection, supporting safer deployment of FL systems.
Blogger's Review: This paper presents practical detection methods for targeted overfitting in federated learning, emphasizing the importance of proactive client defenses. The combination of various detection techniques not only enhances system security but also offers new insights for future privacy protection technologies. Effective detection mechanisms will undoubtedly promote the broader application of federated learning.