My research asks what happens when a federated learning client hides on purpose. Availability is part of the problem, not just a network condition.
The question
Federated aggregation combines updates from clients. On a contested edge network, a client can choose not to participate. My IC3AI paper studies adversary-aware aggregation in this setting.
The paper, Adversary-Aware Aggregation for Federated Learning on Contested Edge Networks, is accepted at IC3AI 2026. The conference is now in April 2027. Acceptance is not the same as publication.
What I am testing
A follow-up study is in preparation. It is pre-registered: I wrote down the questions and the tests before the runs. It re-tests the result across seeds and datasets, and measures which server defences survive an adversary that chooses who goes silent.
The next question
Energy-MNAR takes the availability question to federated load forecasting. When meters drop out for structural reasons, availability can depend on the data itself. That project is designed; it has no results to report here yet.