Federated Neural Decoding

A training paradigm in which multiple sites or users jointly improve a decoder while keeping raw neural recordings local; only model updates are exchanged and combined. For BCI it is attractive because it can pool statistical power across subjects and clinics — helpful for cross-subject foundation models and rare-condition data — without centralizing highly sensitive recordings.

Federation reduces but does not eliminate leakage: shared gradients can still be inverted or subjected to membership inference, so it is usually combined with secure aggregation (the server sees only a sum of updates) and differential privacy. It also inherits robustness problems, since a malicious participant can poison the shared model.