Deep neural network models of emotion understanding

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Affective Science

Background:

• Deep neural networks (DNNs) are increasingly utilized as a computational framework for cognitive modeling.
• Emotion understanding involves complex processes including perception, prediction, and regulation.
• Existing theoretical perspectives in affective science include domain-general (e.g., Bayesian cognitive modeling) and domain-specific (e.g., theory of constructed emotion) frameworks.

Purpose of the Study:

• To provide a focused discussion on the application of DNNs for modeling emotion understanding.
• To define and explore how DNNs can model the key components of emotion understanding: perception, prediction, and regulation.
• To position the contributions of DNN models within existing theoretical frameworks in affective science.

Main Methods:

• Review and discussion of deep learning architectures relevant to emotion understanding.
• Analysis of how different DNN architectures can be applied to model emotion perception, prediction, and regulation.
• Comparison of DNN-based cognitive models with established theoretical perspectives in affective science.

Main Results:

• DNNs can model the perception, prediction, and regulation aspects of emotion understanding using various deep learning architectures.
• DNN models offer a valuable computational lens for affective science, complementing existing theories like Bayesian cognitive modeling and the theory of constructed emotion.
• DNNs present both strengths and limitations as cognitive models for emotion understanding.

Conclusions:

• Deep neural networks provide a powerful computational framework for advancing cognitive models of emotion understanding.
• Guidance is offered on leveraging the strengths of DNNs while mitigating their limitations in affective science research.
• Further integration of DNNs holds promise for a more comprehensive understanding of the computational underpinnings of emotion.