CTNN: A Convolutional Tensor-Train Neural Network for Multi-Task Brainprint Recognition.

Area of Science:

• Neuroscience
• Biometrics
• Artificial Intelligence

Background:

• Brainprint, an EEG-based biometric, offers intrinsic identity linkage.
• Current brainprint recognition relies on traditional machine learning, limited to single cognitive tasks.
• Deep learning excels at feature extraction but requires extensive training data, posing challenges for small-sample, multi-individual scenarios.

Purpose of the Study:

• To propose a novel deep learning model, the Convolutional Tensor-Train Neural Network (CTNN), for multi-task brainprint recognition.
• To address the challenge of limited training samples in realistic brainprint recognition applications.
• To develop a method that efficiently utilizes multi-task learning for brainprint identification.

Main Methods:

• Utilized a Convolutional Neural Network (CNN) with depthwise separable convolution for extracting local temporal and spatial features from EEG data.
• Implemented TensorNet (TN) using low-rank representation to capture multilinear intercorrelations and integrate local information globally.
• Developed the Convolutional Tensor-Train Neural Network (CTNN) model for efficient multi-task brainprint recognition with minimal parameters.

Main Results:

• Achieved high recognition accuracy exceeding 99% across four diverse datasets.
• Demonstrated efficient multi-task brainprint exploitation and effective scaling with limited training samples.
• Identified an interpretable biomarker, highlighting seven specific EEG channels crucial for recognition tasks.

Conclusions:

• The CTNN model offers a robust and accurate solution for brainprint recognition, particularly in scenarios with limited data.
• The proposed method effectively handles multi-task learning for EEG-based biometrics.
• The identification of key EEG channels provides valuable insights into brainprint recognition mechanisms.