392119 Explainable AI for Affective Computing (S) (SoSe 2025)

Affective computing deals systems that process or simulate human affect, including emotions, through the analysis of human behavior. Increasingly applications for the detection of affect and emotions are moving toward the use of deep learning and other black box models. Due to the strongly personal nature of behavior analysis and significant variations in emotional expression between individuals, models should be made transparent and explainable. In this seminar, we will learn about state-of-the-art approaches and associated challenges for explainable affective computing. Since affective computing typical deals with multimodal data, this seminar will look at explainability methods for visual, audio, and text data. Additionally, we will explore emerging research in explainability for multimodal systems. This seminar will include both theoretical and practical methods. Therefore, you will be expected to have a working knowledge of implementing and working with deep learning models in TensorFlow and Keras.