Synthesizing affective neurophysiological signals using generative models: A review paper

Area of Science:

• Neuroscience and Artificial Intelligence
• Human-Computer Interaction
• Affective Computing

Background:

• Advancing human-computer interaction requires integrating emotional intelligence into machines.
• Reliable emotion recognition systems are crucial for this integration.
• A significant challenge is the scarcity of public affective datasets for training these systems.

Purpose of the Study:

• To review the application of generative models for augmenting neurophysiological datasets (EEG, fNIRS) in emotion recognition.
• To analyze different generative models, their input, deployment, and data synthesis evaluation.
• To provide insights into the benefits, challenges, and future directions of using generative models in this field.

Main Methods:

• Comprehensive literature review focusing on generative models for neurophysiological signals.
• Analysis of input formulation, deployment strategies, and quality evaluation of synthesized data.
• Examination of models applied to Electroencephalogram (EEG) and Functional Near-Infrared Spectroscopy (fNIRS) data.

Main Results:

• Generative models offer a viable solution to the data scarcity problem in affective computing.
• Various generative approaches exist, each with specific strengths and weaknesses in data synthesis.
• Methodologies for evaluating the quality of generated neurophysiological data are crucial for reliable system development.

Conclusions:

• Generative models are essential for advancing neurophysiological data augmentation in emotion recognition.
• Addressing challenges in model application and evaluation will lead to more robust and efficient systems.
• This review facilitates progress in developing better human-computer interaction through enhanced emotion recognition.