Layer-wise Interpretability Investigation of Facial Expression Recognition Models Based on Grad-CAM

DOI

10.2991/978-94-6463-540-9_102How to use a DOI?

Keywords

Convolutional Neural Networks; Facial Expression Recognition; Gradient-weighted Class Activation Mapping

Abstract

For a long time, artificial intelligence has faced the challenge of interpretability, with the black-box problem persistently troubling researchers. Although there have been studies using Gradient-weighted Class Activation Map (Grad-CAM) for interpretability in the field of facial expression recognition, these studies often lack attention to the impact of each layer of the model on interpretability. In this study, different models are constructed, and Grad-CAM is used to explore the impact of different types of layers on model interpretability, filling the gap left by previous work. To be more specifically, this study constructed various Convolutional Neural Networks (CNN) models, including a baseline model, three models with modified convolutional layers, and three models with modified pooling layers. For comparative experiments, all six modify models were modified from the baseline model. All these models were trained using the FER-2013 dataset. Before training, the dataset underwent image pre-process and augmentation to prevent overfitting. After training these models, Grad-MAPs are generated based on the same test images. Experimental results show that different layers significantly impact model interpretability: convolutional layers affect the size of hotspot regions, while pooling layers influence the discreteness of these regions.

Copyright

© 2024 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Siyuan Yao
PY  - 2024
DA  - 2024/10/16
TI  - Layer-wise Interpretability Investigation of Facial Expression Recognition Models Based on Grad-CAM
BT  - Proceedings of the 2024 2nd International Conference on Image, Algorithms and Artificial Intelligence (ICIAAI 2024)
PB  - Atlantis Press
SP  - 1026
EP  - 1032
SN  - 2352-538X
UR  - https://doi.org/10.2991/978-94-6463-540-9_102
DO  - 10.2991/978-94-6463-540-9_102
ID  - Yao2024
ER  -