TASA: Temporal Attention With Spatial Autoencoder Network for Odor-Induced Emotion Classification Using EEG.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Affective Computing

Background:

• The human olfactory system links smell with emotions.
• Electroencephalogram (EEG) offers non-invasive, high temporal resolution for studying odor perception.
• Accurate analysis of EEG's spatial and temporal features is key to understanding odor-induced emotional valence.

Purpose of the Study:

• To propose a novel deep learning architecture, Temporal Attention with Spatial Autoencoder Network (TASA), for predicting odor-induced emotions from EEG.
• To enhance the learning of spatial information using an autoencoder for reduced data loss.
• To effectively model the temporal dynamics of olfactory responses using Long Short-Term Memory with Multi-Head Self-Attention (LSTM-MSA).

Main Methods:

• Developed TASA, a deep learning model incorporating a filter-bank, spatial encoder, time segmentation, LSTM, and Multi-Head Self-Attention (MSA) layers.
• Employed a two-phase learning framework: spatial information learning via autoencoder reconstruction and temporal dynamics learning via LSTM-MSA.
• Evaluated TASA on an existing olfactory EEG dataset, comparing its performance against established deep learning architectures.

Main Results:

• TASA demonstrated superior effectiveness in predicting olfactory-triggered emotional responses compared to existing deep learning models.
• The autoencoder module successfully learned spatial electrode information, minimizing data loss.
• The LSTM-MSA module effectively captured temporal dynamics crucial for olfactory processing.

Conclusions:

• TASA provides a robust framework for analyzing EEG data to recognize olfactory-induced emotions.
• The model's interpretability analysis confirmed its ability to learn relevant spatial-spectral features.
• This research advances objective methods for studying the complex relationship between smell and emotion.