Interactive EEG Emotion Recognition with Incremental Gaussian Processes.

Area of Science:

• Neuroscience
• Machine Learning
• Human-Computer Interaction

Background:

• Existing electroencephalogram (EEG)-based emotion recognition models lack interactivity and struggle with prediction uncertainty.
• Static training paradigms limit the adaptability and performance optimization of current EEG emotion recognition systems.

Purpose of the Study:

• To develop a novel paradigm for interactive emotion recognition using incremental Gaussian processes (GP).
• To enhance EEG-based emotion recognition by enabling models to adjust based on user feedback and manage prediction uncertainty.

Main Methods:

• Utilizing a Gaussian process (GP) framework to model emotion recognition and quantify prediction uncertainty via variance.
• Implementing an expert interaction mechanism for targeted sample correction based on high uncertainty.
• Developing an incremental update strategy for efficient model refinement without reprocessing all data.
• Employing sparse approximation and variational inference to manage the computational complexity of GPs.

Main Results:

• The proposed interactive GP method demonstrated significant advantages over state-of-the-art (SOTA) methods in both subject-dependent and subject-independent experiments.
• Achieved highest improvement in Dominance (1.73%) on the DREAMER dataset (subject-dependent).
• Attained largest performance improvement in Arousal (2.96%) on the DEAP dataset (subject-independent).

Conclusions:

• The novel interactive paradigm effectively enhances EEG-based emotion recognition by integrating expert feedback.
• The incremental GP approach offers an efficient and effective solution for handling uncertainty and optimizing model performance.
• This method represents a significant advancement in creating more adaptive and accurate emotion recognition systems.