Different Set Domain Adaptation for Brain-Computer Interfaces: A Label Alignment Approach.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interface (BCI) systems require extensive calibration, hindering real-world use.
• Domain adaptation techniques leverage existing data to reduce calibration needs.
• Existing methods often assume identical feature and label spaces between source and target domains.

Purpose of the Study:

• To address the challenge of different label spaces in BCI domain adaptation.
• To introduce a practical setting and a novel approach for cross-label BCI adaptation.
• To reduce the calibration effort for BCI users.

Main Methods:

• Proposed a novel label alignment (LA) approach for BCI domain adaptation.
• LA aligns source and target label spaces with different sets.
• LA requires minimal labeled data from the target subject (one sample per class).

Main Results:

• The proposed LA method effectively aligns different label spaces in BCIs.
• LA can be used as a preprocessing step for various feature extraction and classification algorithms.
• LA demonstrated effectiveness in experiments on two motor imagery datasets.

Conclusions:

• The novel label alignment (LA) approach successfully handles different label sets in BCI domain adaptation.
• LA significantly reduces the need for extensive subject-specific calibration.
• LA offers a flexible and effective solution for improving BCI system adaptability.