Double-Criteria Active Learning for Multiclass Brain-Computer Interfaces

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Technological advancements allow collection of large electroencephalography (EEG) datasets.
• Labeled EEG data is costly and time-consuming to acquire for brain-computer interface (BCI) systems.

Purpose of the Study:

• To propose a novel active learning method to minimize labeled, subject-specific EEG data for effective classifier training.
• To combine uncertainty and representativeness measures within an extreme learning machine (ELM) for efficient BCI development.

Main Methods:

• Utilized an extreme learning machine (ELM) classifier to select unlabeled examples based on uncertainty (best-versus-second-best strategy).
• Measured sample diversity against labeled data and similarity among unlabeled samples.
• Introduced a tradeoff parameter to balance informative and representative samples for classifier construction.

Main Results:

• The proposed active learning method demonstrated superior or comparable performance to state-of-the-art algorithms on benchmark and multiclass motor imagery EEG datasets.
• Experimental results confirmed the method's efficacy in improving classifier performance.
• The approach significantly reduced the requirement for training samples in BCI applications.

Conclusions:

• The novel active learning strategy effectively enhances classifier performance in BCI systems.
• The method substantially decreases the amount of labeled EEG data needed, addressing cost and time constraints.
• This approach offers a more efficient pathway for developing robust BCI systems.