TriNet-MTL: A Multi-Branch Deep Learning Framework for Biometric Identification and Cognitive State Inference from...

Area of Science:

• Neuroscience and Artificial Intelligence
• Brain-Computer Interfaces
• Biometrics and Cognitive Science

Background:

• Auditory-evoked electroencephalography (EEG) signals contain rich temporal and cognitive features for individual identification and neural response analysis.
• Traditional unimodal approaches inadequately leverage multidimensional EEG information, limiting real-world applications in biometrics and neurocognition.
• There is a need for unified models to exploit both physiological and cognitive dimensions of auditory-evoked EEG.

Purpose of the Study:

• To develop a unified deep learning model for joint biometric identification, auditory stimulus language classification, and device modality recognition using auditory-evoked EEG.
• To introduce TriNet-MTL (Triple-Task Neural Transformer for Multitask Learning), a novel deep learning framework for integrated EEG analysis.
• To exploit both user-specific physiological and cognitively informative patterns within auditory-evoked EEG signals.

Main Methods:

• Development of TriNet-MTL, a multi-branch deep learning framework with a shared temporal encoder and transformer-based sequence modeling.
• Training and validation on auditory-evoked EEG data from 20 human participants.
• Simultaneous learning of task-specific features via three output heads for user identity, stimulus language, and stimulus delivery mode, optimized with joint cross-entropy loss.

Main Results:

• TriNet-MTL demonstrated robust performance across all three classification tasks.
• High accuracy (>93%) achieved in biometric identification.
• Strong generalization in cognitive state inference, with multitask training enhancing representation learning and task synergy.

Conclusions:

• The TriNet-MTL framework effectively captures user-specific and cognitively informative patterns from auditory-evoked EEG.
• This approach establishes a promising direction for integrated EEG-based biometric authentication and cognitive state monitoring.
• The study highlights the potential of multitask learning in EEG analysis for enhanced accuracy and reliability in real-world systems.