Complex machine learning models such as deep convolutional neural networks and recursive neural networks have recently made great progress in a wide range of computer vision applications, such as object/scene recognition, image captioning, visual question answering. But they are often perceived as black-boxes. As the models are going deeper in search of better recognition accuracy, it becomes even harder to understand the predictions given by the models and why.

Continuing from the 1st Tutorial on Interpretable Machine Learning for Computer Vision at CVPR’18, the 2nd Tutorial at ICCV’19, and the 3rd Tutorial at CVPR’20 where more than 1000 audiences attended, this series tutorial is designed to broadly engage the computer vision community with the topic of interpretability and explainability in computer vision models. We will review the recent progress we made on visualization, interpretation, and explanation methodologies for analyzing both the data and the models in computer vision. The main theme of the tutorial is to build up consensus on the emerging topic of machine learning interpretability, by clarifying the motivation, the typical methodologies, the prospective trends, and the potential industrial applications of the resulting interpretability.