Personalised federated learning improves eeg decoding accuracy and efficiency
Personalised federated learning for Riemannian and Euclidean EEG decoding
Machine Learning
Summary
Decoding brain signals like EEG is hard because everyone's brain signals look a bit different. The researchers studied a way to train brain-signal reading models across many people without sharing their data directly, which keeps privacy intact. They improved this approach by letting the model share common knowledge but customize parts for each person. This method worked better and faster than past approaches on several brain signal datasets, especially when using a special mathematical model called SPDNet.
What this means in practice
- •For medical device developers: Build more accurate and efficient brain-computer interfaces that respect user privacy by personalising federated learning models for EEG decoding.$Commercial implications: Enables selling advanced brain-computer interface products personalized to individuals without sharing sensitive brain data.
- •For signal processing engineers: Implement personalised federated learning frameworks that reduce communication costs and training time in multi-subject EEG decoding tasks.
Authors
Thibault Pautrel, Florent Bouchard, Ammar Mian, Guillaume Ginolhac
Abstract
Federated learning (FL) lets EEG decoders learn from recordings of several subjects without pooling them. We consider two light EEG decoders, the Riemannian SPDNet and the Euclidean EEGNet. Both split into a trunk, which builds a latent representation, and a head, which classifies it. Inter-subject variability, however, makes a single shared FL model a poor fit for each subject. Personalised FL addresses this: all subjects learn a common trunk, and each subject keeps its own head. We adapt it for SPDNet and study its effects against standard FL and centralised training, with EEGNet as a Euclidean baseline. Experiments cover three motor-imagery datasets that span diverse regimes in channels, subjects and classes. We observe that personalised SPDNet reaches higher accuracy than both standard FL and centralised training, while converging in fewer rounds and communicating fewer parameters than standard FL. It also outperforms every EEGNet configuration on two of the three datasets, although centralised EEGNet outperforms centralised SPDNet.