Deep neural network models of emotion understanding

Deep neural networks provide a useful computational framework for constructing cognitive models of emotion understanding. This paper introduces readers to this modeling approach. It begins by providing a brief primer on artificial neural networks. It then discusses the advantages of using deep learning to model the mind and brain – particularly the flexibility and scalability of these models. Next, it considers the limitations of deep neural networks as cognitive models, including their biological (im)plausibility, difficulty solving compositional problems, and challenges to explainability. Mitigation measures for each of these limitations are discussed. The paper then focuses on how to match neural network architectures with different components of emotion understanding. Three different components are discussed: emotion perception, emotion prediction, and (social) emotion regulation. The paper closes by considering applications and future directions of deep learning models of emotion understanding.