Label-revealed online updates improve brain signal decoding accuracy
Benchmarking Label-Revealed Online Updates for EEG BCI Decoding
Machine Learning
Summary
Brain-computer interfaces (BCIs) interpret brain signals, but over time, these signals can change and make the BCIs less accurate. The authors study two popular methods for analyzing brain signals and test how updating the models with new information as data arrives helps improve accuracy. They found that using label-revealed updates—knowing the correct answer after each prediction—makes the models better on most tested datasets. They also showed that recent data is more valuable for updating the model, but older data still helps a little.
What this means in practice
- •For bci device developers: Improve real-time decoding accuracy of EEG signals by integrating label-revealed online model updating.$Commercial implications: Enables more accurate and adaptive BCIs for medical or assistive technology markets by reducing model degradation over time.
- •For signal processing teams: Enhance robustness of motor-imagery EEG decoding by balancing incorporating new data and forgetting older recordings.
Authors
Bogdan Kozyrskiy, Artem Grachev, Abraham I. Camelo Guerrero
Abstract
Electroencephalography (EEG) signals drift over time, which can cause static brain-computer interface (BCI) models to degrade in practice. We present a benchmark for online adaptation and compare two widely used pipeline families, Common Spatial Patterns (CSP) and Riemannian covariance-based methods, under time-ordered prequential (test-then-train) evaluation. We examine (i) which pipelines benefit most from label-revealed updates, (ii) whether controlled forgetting of older data improves robustness, and (iii) how a minimal-calibration cold start compares with starting from a pretrained model. Across four datasets (three motor-imagery datasets and one movement-decoding dataset), label-revealed online updates improve 13 of 14 model/dataset pairs on the two largest streams, with relative accuracy gains of up to about 18% over a frozen model. A Shapley-based data-valuation analysis over temporal blocks assigns the largest mean value to the most recent block in each of the three analyzed datasets, while older blocks retain positive value.