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

Abstract

Electroencephalography (EEG) headsets are the most commonly used sensing devices for brain-computer interface. In real-world applications, there are advantages to extrapolating data from one user session to another. However, these advantages are limited if the data arise from different hardware systems, which often vary between application spaces. Currently, this creates a need to recalibrate classifiers, which negatively affects people's interest in using such systems. In this paper, we employ active weighted adaptation regularization (AwAR), which integrates weighted adaptation regularization (wAR) and active learning, to expedite the calibration process. wAR makes use of labeled data from the previous headset and handles class-imbalance, and active learning selects the most informative samples from the new headset to label. Experiments on single-trial event-related potential classification show that AwAR can significantly increase the classification accuracy, given the same number of labeled samples from the new headset. In other words, AwAR can effectively reduce the number of labeled samples required from the new headset, given a desired classification accuracy, suggesting value in collating data for use in wide scale transfer-learning applications.

Online and Offline Domain Adaptation for Reducing BCI Calibration Effort

“…7) wARSDS and wAR had very similar performance, but instead of using 13 auxiliary subjects, wARSDS only used on average 6.84 subjects for BioSemi, 6.03 subjects for ABM, and 6.85 subjects for Emotiv, corresponding to 47.38%, 53.62% and 47.31% computational cost saving, respectively. As in [51], [53], we also performed comprehensive statistical tests to check if the performance differences among the six algorithms (BL1 was not included because it is not iterative) were statistically significant. We used the area-underperformance-curve (AUPC) [31], [51], [53] to assess overall performance differences among these algorithms.…”

“…This section describes the offline wARSDS algorithm [51], [53], which originates from the adaptation regularizationregularized least squares (ARRLS) algorithm in [25]. We made several major enhancements to ARRLS to better handle classimbalance and multiple source domains, and also to make use of labeled samples in the target domain.…”

“…Our future research will consider the more general case that P s (y) and P t (y) are different. Let the classifier be f (x) = w T φ(x), where w is the classifier parameters, and φ : X → H is the feature mapping function that projects the original feature vector to a Hilbert space H. As in [51], [53], the learning framework of wAR is formulated as:…”

“…The optimal RBF parameter was found by cross-validation. We chose w t = 2, σ = 0.1, and λ = 10, following the practice in [25], [51], [53].…”

“…An active TL approach was proposed in [48] for offline calibration, which combined TL and active learning to minimize the offline calibration effort. An active weighted adaptation regularization approach, which combines active learning, TL, regularization and semi-supervised learning, was proposed in [51] to facilitate the switching between different EEG headsets. All these approaches are for BCI classification problems.…”