Brain Emotion Perception Inspired EEG Emotion Recognition With Deep Reinforcement Learning.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• The Papez circuit theory and reinforcement learning principles inform brain emotion mechanisms.
• Electroencephalogram (EEG) signals offer insights into cognitive and emotional processes.
• Deep Q networks (DQN) are effective in reinforcement learning tasks.

Purpose of the Study:

• To develop a novel deep Q network model integrating frontal lobe EEG signals for enhanced emotion recognition.
• To simulate brain emotion mechanisms, particularly the Papez circuit, within an artificial intelligence framework.
• To improve the accuracy of predicting valence and arousal dimensions in emotion perception.

Main Methods:

• A double dueling deep Q network (DQN) was designed, named frontal lobe double dueling DQN (FLD3QN).
• The FLD3QN framework incorporates frontal lobe EEG signals as prior information.
• A bifrontal lobe residual convolution neural network (BiFRCNN) simulated parts of the Papez circuit, and a step penalty factor was implemented.

Main Results:

• Ablation studies on the DEAP dataset demonstrated the effectiveness of the FLD3QN model.
• The model showed significant increases in average accuracies for valence (25.24%) and arousal (23.31%) dimensions.
• The study verified the crucial roles of the frontal lobe and Papez circuit in emotion perception learning.

Conclusions:

• The frontal lobe double dueling DQN (FLD3QN) effectively models emotion perception and reward learning.
• Integrating frontal lobe EEG signals and simulating brain circuits enhances emotion recognition capabilities.
• The developed model offers a promising approach for advancing affective computing and brain-computer interfaces.