| Creators: |
Gebele, Jens and Brune, Philipp and Schwab, Frank and von Mammen, Sebastian |
| Title: |
Interpreting Emotions Through the Grad-CAM Lens: Insights and Implications in CNN-Based Facial Emotion Recognition |
| Item Type: |
Conference or Workshop Item |
| Event Title: |
(Proceedings of the) 27th International Conference, ICPR "Pattern Recognition" ; Part XIII |
| Event Location: |
Kolkata, India |
| Event Dates: |
December 1–5, 2024 |
| Projects: |
TTZ-GZ |
| Page Range: |
pp. 414-429 |
| Date: |
2 December 2024 |
| Divisions: |
Informationsmanagement |
| Abstract (ENG): |
Facial Emotion Recognition (FER) has gained significant attention in recent years due to its potential applications in various fields such as automotive, mental health, and education. Despite the impressive results of Deep Learning (DL) in these areas, a critical shortfall of these systems is the lack of explainability. This paper presents a systematic analysis using Gradient-weighted Class Activation Mapping (Grad-CAM) combined with Guided Backpropagation to investigate the features learned by DL models in FER and their alignment with established emotional theories. We apply this methodology to Convolutional Neural Networks (CNNs) trained on three different emotional datasets: FER-2013, RAF-DB, and AffectNet. Our findings indicate that machine-learned features vary within an emotional state and are not necessarily aligned with expert understanding in emotion psychology. This raises questions about the reliability and ethical implications of using FER systems in sensitive areas, where accurate interpretation of emotions is critical. In response, our study proposes exploring Neuro-Symbolic AI approaches as a potential pathway to more effectively grasp the complexity of emotion psychology and address these concerns. This approach paves the way for the development of new FER model architectures, potentially fostering the emergence of more nuanced emotional concepts. |
| Forthcoming: |
No |
| Language: |
English |
| Uncontrolled Keywords: |
Grad-CAM, Guided Grad-CAM, Explainable AI, Emotion Recognition, Facial Emotion Recognition, Convolutional Neural Network |
| Citation: |
Gebele, Jens and Brune, Philipp and Schwab, Frank and von Mammen, Sebastian
(2024)
Interpreting Emotions Through the Grad-CAM Lens: Insights and Implications in CNN-Based Facial Emotion Recognition.
In: (Proceedings of the) 27th International Conference, ICPR "Pattern Recognition" ; Part XIII, December 1–5, 2024, Kolkata, India, pp. 414-429.
(Lecture Notes in Computer Science; 15313).
ISBN 9783031782015
|
