Transfer learning with large-scale data in brain-computer interfaces.

Area of Science:

• Neuroscience
• Computer Science
• Human-Computer Interaction

Background:

• Human variability in electroencephalogram (EEG) data presents a major obstacle for practical brain-computer interface (BCI) development.
• Collecting extensive user-specific training data is time-consuming and labor-intensive, limiting BCI usability.

Purpose of the Study:

• To propose a novel transfer learning (TL) framework to minimize the requirement for subject-specific calibration data in BCIs.
• To validate the efficacy of this TL framework in a passive BCI for detecting neurocognitive lapses during driving.

Main Methods:

• Leveraging large-scale datasets from multiple subjects to train a TL model.
• Implementing a passive BCI system designed to identify lapses in cognitive function while driving.
• Comparing the performance of the proposed TL approach against traditional within-subject methods.

Main Results:

• The novel TL framework significantly outperformed the within-subject approach.
• The proposed method drastically reduced the calibration data needed per individual to approximately 1.5 minutes.
• This reduction contrasts sharply with the 90 minutes typically required for standard within-subject calibration.

Conclusions:

• The developed TL framework effectively minimizes the need for subject-specific calibration data.
• This advancement has the potential to greatly accelerate the adoption and real-world application of BCIs.
• The study demonstrates a viable solution for overcoming data acquisition challenges in BCI development.