Dimensionality Transcending: A Method for Merging BCI Datasets With Different Dimensionalities.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCI) often rely on electroencephalography (EEG) data.
• EEG datasets can vary significantly due to different experimental setups, such as electrode configurations, limiting data transferability.
• Symmetric Positive Definite (SPD) matrices are used to represent the statistical properties of EEG signals.

Purpose of the Study:

• To develop a transfer learning method capable of harmonizing EEG datasets with varying dimensionalities and statistical distributions.
• To enable the transfer of discriminative information between BCI systems that originate from different experimental conditions.
• To address the challenge of data incompatibility in BCI research arising from diverse experimental setups.

Main Methods:

• A two-step procedure involving isometric transformations for dimensionality matching.
• Domain adaptation techniques tailored to the intrinsic geometry of SPD matrices for statistical distribution matching.
• Application of the method to time-series EEG data from BCI systems with differing electrode placements and counts.

Main Results:

• Demonstrated successful transfer of discriminative information between BCI recordings from disparate experimental setups.
• The proposed method effectively matched datasets in terms of dimensionality and statistical distribution.
• EEG data from BCI systems with different electrode configurations were rendered compatible for information transfer.

Conclusions:

• The developed transfer learning method facilitates the reuse of information across diverse BCI datasets.
• This approach paves the way for more robust and adaptable BCI systems that can learn from multiple data sources.
• Overcoming experimental setup differences enhances the potential for creating next-generation BCI technologies.