Latent Factor Decoding of Multi-Channel EEG for Emotion Recognition Through Autoencoder-Like Neural Networks

Area of Science:

• Neuroscience
• Machine Learning
• Signal Processing

Background:

• Cross-subject emotion recognition from multichannel EEG is challenging.
• Identifying consistent brain variables across individuals during emotional processing is hypothesized to improve model robustness.

Purpose of the Study:

• To develop a robust emotion recognition framework using unsupervised deep generative models for multichannel EEG.
• To explore the utility of variational autoencoders (VAEs) for decoding latent emotional states from EEG data.

Main Methods:

• Utilized a variational autoencoder (VAE) to learn latent factors from multichannel EEG data.
• Employed sequence modeling techniques to evaluate emotion recognition performance based on the learned latent factors.
• Validated the methodology on the DEAP and SEED public datasets.

Main Results:

• Autoencoder-like neural networks demonstrated suitability for unsupervised EEG modeling.
• The proposed framework achieved inspiring performance in emotion recognition.
• The VAE-based approach outperformed traditional methods like ICA and standard autoencoders in this context.

Conclusions:

• This work presents the first application of VAEs for multichannel EEG decoding in emotion recognition.
• The proposed method offers a feasible and promising approach for emotion recognition.
• The framework holds potential for diagnosing neurological conditions like depression, Alzheimer's disease, and mild cognitive impairment.