Siamese Neural Networks for EEG-based Brain-computer Interfaces.

Area of Science:

• Neuroscience and Artificial Intelligence
• Brain-Computer Interface (BCI) Technology
• Signal Processing and Machine Learning

Background:

• Brain-computer interfaces (BCI) aim to bridge human brain communication with computers.
• Electroencephalogram (EEG) is a primary method for monitoring brain electrical activity in BCI.
• Current BCI systems struggle with performance degradation as the number of mental tasks increases.

Purpose of the Study:

• To develop a scalable EEG processing and feature extraction paradigm for multi-class BCI problems.
• To enhance the performance of BCI systems by addressing the limitations of existing classification methods.

Main Methods:

• Proposed a novel Siamese neural network architecture based on Convolutional Neural Networks (CNN).
• Employed a contrastive loss function for training the network to verify similarity between EEG trials.
• Integrated the Siamese CNN with One vs. Rest (OVR) and One vs. One (OVO) strategies for multi-class classification.

Main Results:

• The proposed Siamese network architecture demonstrated promising performance in classifying EEG signals.
• Evaluation on a 4-class Motor Imagery (MI) dataset (BCI Competition IV2a) showed competitive results.
• The approach effectively scales for multi-class problems, overcoming a key bottleneck in BCI development.

Conclusions:

• The developed Siamese neural network paradigm offers an effective solution for multi-class EEG-based BCI.
• This approach represents a significant advancement in improving the practical applicability and scalability of BCI systems.
• Further research can explore the integration of this method into more complex BCI applications.