A Transfer Learning SSVEP Decoding Algorithm Calibrated With Single-Trial Data

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Training-based algorithms excel in SSVEP-BCIs but require extensive calibration data.
• Calibration demands limit BCI practicality due to user fatigue and cost.
• Existing transfer learning methods often need substantial source or target domain data.

Purpose of the Study:

• To introduce cross-dataset transfer learning for SSVEP BCIs.
• To address the data mismatch problem in cross-dataset transfer learning.
• To develop a practical SSVEP decoding algorithm with minimal calibration.

Main Methods:

• Proposed a novel cross-dataset transfer learning approach for SSVEP.
• Introduced TL-CSTD (Transfer Learning for SSVEP decoding calibrated with Single-Trial data).
• Utilized 2s of single-trial calibration data for template matching and knowledge extraction.

Main Results:

• TL-CSTD effectively overcomes the data mismatch problem.
• Achieved excellent SSVEP recognition performance with only 2s of calibration data.
• Demonstrated effectiveness across three large SSVEP datasets.

Conclusions:

• TL-CSTD significantly enhances the practicality and application potential of SSVEP-BCIs.
• The method reduces the need for extensive user training and calibration.
• This approach offers a viable solution for efficient and user-friendly BCI systems.