Lightweight deep learning models for EEG decoding: a review.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-computer interface (BCI) technology utilizes electroencephalography (EEG) signals for device control.
• Deep learning models excel at EEG signal classification but often have high computational demands.
• Lightweight deep learning architectures are crucial for real-time and portable BCI systems.

Purpose of the Study:

• To systematically review lightweight deep learning models for EEG signal classification.
• To categorize existing approaches into distinct strategies for clarity.
• To identify trends and future research directions in efficient BCI model design.

Main Methods:

• Categorization of lightweight deep learning models into three main strategies: information integration, hidden layer optimization, and hybrid structural optimization.
• Systematic review of recent advancements in each category.
• Analysis of model efficiency and performance trade-offs.

Main Results:

• Deep learning significantly improves EEG classification accuracy over traditional methods.
• Lightweight models address computational and memory limitations of complex deep learning architectures.
• Three primary strategies exist for developing efficient EEG classification models.

Conclusions:

• Lightweight deep learning models are essential for practical, real-world BCI applications.
• Further research into optimized architectures will enhance BCI usability and accessibility.
• Efficient EEG classification is key to advancing neurorehabilitation and assistive technologies.