The presented paper entitled "Interpreting Emotions through the Grad-CAM Lens: Insights and Implications in CNN-Based Facial Emotion Recognition" is dedicated to the explainability of AI systems in the field of Facial Emotion Recognition (FER). FER is increasingly being improved through the use of deep learning (DL) methods. DL is based on artificial neural networks that are capable of recognising complex and non-linear relationships and has applications in education, mental health and the automotive industry.
However, the study shows that the features learnt by DL models often do not correspond to psychological theories from emotion research, which raises ethical and methodological questions. As a forward-looking solution, the paper proposes the use of Neuro-Symbolic AI, which combines classical AI methods with modern DL approaches to better reflect the complexity of emotional concepts and develop more reliable models.
The presentation at the ICPR, a platform for international scientific exchange, not only emphasises the relevance and quality of AI research at TTZ Günzburg and HNU, but also provides valuable impetus for practical application in the region and beyond.
Contact person: