Suppression of negative transfer in motor imagery brain-computer interface based on mutual information and Pearson...

Area of Science:

• Neuroscience
• Computer Science
• Machine Learning

Background:

• Brain-computer interfaces (BCIs) face challenges in transfer learning due to individual differences.
• Data characteristic length and source domain sample selection significantly impact BCI performance.

Purpose of the Study:

• To address negative migration caused by feature length in BCI transfer learning.
• To improve classification accuracy and learning efficiency in BCIs.
• To develop a method for optimal source domain participant selection.

Main Methods:

• Mutual Information Transfer (MIT) algorithm to select effective features based on entropy.
• Pearson Correlation Coefficient Source Domain Automatic Selection (PDAS) algorithm to select appropriate participants.
• Offline and online testing on public datasets, comparing with existing algorithms.

Main Results:

• The MIT algorithm effectively reduces negative migration and improves learning efficiency.
• The PDAS algorithm automatically selects suitable source domain participants, enhancing accuracy.
• Combined MIT + PDAS algorithms demonstrate significant advantages over existing methods.

Conclusions:

• The proposed MIT and MIT + PDAS algorithms offer substantial improvements for BCI transfer learning.
• These methods effectively mitigate challenges related to data characteristics and individual differences.
• The algorithms show promise for more accurate and efficient BCI systems.