A Survey on Deep Learning-Based Short/Zero-Calibration Approaches for EEG-Based Brain-Computer Interfaces.

Area of Science:

• Neuroscience and Artificial Intelligence
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) offer a communication pathway between users and external systems.
• Electroencephalography (EEG) is a practical but data-intensive method for BCIs, often requiring lengthy calibration.
• Deep learning (DL) shows promise in analyzing complex EEG patterns and reducing calibration needs.

Purpose of the Study:

• To review deep learning-based methods for short/zero-calibration in BCIs.
• To explore methodological and algorithmic trends in DL for BCI calibration.
• To discuss future research directions in DL-based BCI calibration.

Main Methods:

• Review of deep learning techniques for BCI calibration, focusing on data augmentation (DA) and transfer learning (TL).
• Categorization of DA methods into generative model-based and geometric manipulation-based approaches.
• Classification of TL techniques into explicit and implicit methods.

Main Results:

• Deep learning significantly facilitates short/zero-calibration in EEG-based BCIs, reducing data acquisition burdens.
• ~45% of reviewed DA studies utilized generative model-based techniques.
• ~45% of reviewed TL studies employed explicit knowledge transfer strategies.

Conclusions:

• Advances in DA and TL are crucial for efficient DL-based BCIs.
• Generative models and explicit TL strategies are prominent in current research.
• Recommendations are provided for DA strategies and trends in TL for DL-based BCIs.