Switching EEG Headsets Made Easy: Reducing Offline Calibration Effort Using Active Weighted Adaptation...

Area of Science:

• Neuroscience
• Computer Science
• Machine Learning

Background:

• Electroencephalography (EEG) headsets are primary for brain-computer interfaces (BCI).
• Data transferability between different EEG hardware systems is challenging, necessitating recalibration and hindering user adoption.
• Current limitations in cross-hardware calibration impede the widespread application of BCI technology.

Purpose of the Study:

• To develop a method that expedites the calibration process for BCIs using different hardware.
• To enhance the accuracy and efficiency of BCI classifiers when transferring data across varying EEG systems.
• To reduce the dependency on extensive labeled data from new hardware for BCI recalibration.

Main Methods:

• Active weighted adaptation regularization (AwAR) was employed, integrating weighted adaptation regularization (wAR) and active learning.
• wAR utilized labeled data from a previous headset and addressed class imbalance.
• Active learning identified the most informative samples from a new headset for labeling.

Main Results:

• AwAR significantly increased classification accuracy in single-trial event-related potential classification with a fixed number of labeled samples.
• The method effectively reduced the quantity of labeled data required from a new headset to achieve desired classification accuracy.
• Experiments demonstrated the efficacy of AwAR in overcoming hardware variability challenges in EEG-based BCIs.

Conclusions:

• AwAR offers a promising solution for efficient BCI recalibration across diverse hardware setups.
• The approach facilitates data collation for large-scale transfer-learning applications in BCI.
• By reducing calibration burden, AwAR can increase user interest and adoption of BCI systems.