Relation Learning Using Temporal Episodes for Motor Imagery Brain-Computer Interfaces.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCI) rely on accurate motor imagery (MI) models.
• Collecting extensive MI data is time-consuming and costly for individual users.
• Developing efficient MI models with limited data is crucial for practical BCI applications.

Purpose of the Study:

• To propose a few-shot learning method, Temporal Episode Relation Learning (TERL), for subject-specific MI-BCI.
• To enable reliable MI modeling with minimal trials from a target subject.
• To enhance user experience and facilitate real-world MI-BCI deployment.

Main Methods:

• Developed Temporal Episode Relation Learning (TERL), a novel few-shot learning approach.
• TERL compares MI trials using episode-based training, encoding temporal patterns.
• Evaluated TERL through offline analysis and online simulation on four public MI-BCI datasets.

Main Results:

• TERL demonstrated superior performance compared to baseline and state-of-the-art methods.
• The method achieved competitive results in subject-specific MI-BCI scenarios with limited target subject trials.
• TERL effectively utilizes data from source subjects to aid target subject modeling.

Conclusions:

• TERL offers an effective solution for creating reliable MI-BCI models with few trials.
• The proposed method significantly improves classification performance by incorporating temporal dynamics.
• TERL shows strong potential for practical, user-friendly MI-BCI systems.