M³CV: A multi-subject, multi-session, and multi-task database for EEG-based biometrics challenge.

Area of Science:

• Neuroscience
• Machine Learning
• Biometrics

Background:

• Existing electroencephalography (EEG) studies primarily focus on group-level commonalities, often overlooking significant intra- and inter-subject variability.
• Advances in deep learning offer potential for EEG signal applications, yet challenges persist in cross-session, cross-task, and cross-subject decoding.

Purpose of the Study:

• To launch an EEG-based biometric competition utilizing the large-scale M3CV database.
• To characterize and leverage EEG signal variability across subjects, sessions, and tasks.
• To promote the development of advanced machine learning algorithms for EEG decoding.

Main Methods:

• Development of the M3CV (Multi-subject, Multi-session, and Multi-task) database containing EEG signals from 106 subjects.
• Recording of EEG data across 6 paradigms (resting-state, sensory, cognitive, motor) with 14 signal types and 120,000 epochs.
• Introduction of identification and verification learning tasks, performance metrics, and baseline methods for the competition.

Main Results:

• The M3CV database provides a comprehensive resource for studying EEG commonality and variability.
• The competition framework facilitates the evaluation and advancement of machine learning algorithms for EEG analysis.
• Initial findings highlight the potential for improved EEG decoding by addressing signal variability.

Conclusions:

• The M3CV dataset and associated competition offer a unique platform for in-depth understanding of EEG signal characteristics.
• This initiative aims to drive innovation in machine learning for robust and generalizable EEG-based applications.
• Future research will focus on developing sophisticated algorithms to harness both commonality and variability in EEG data.