Due to impressive performance, deep neural networks for object detection in images have become a prevalent choice. Given the complexity of the neural network models used, users of these algorithms are typically given no hint as to how the objects were found. It remains, for example, unclear whether an object is detected based on what it looks like or based on the context in which it is located. We have developed an algorithm, Surrogate Object Detection Explainer (SODEx), that can explain any object detection algorithm using any classification explainer. We evaluate SODEx qualitatively and quantitatively by detecting objects in the COCO dataset with YOLOv4 and explaining these detections with LIME. This empirical evaluation does not only demonstrate the value of explainable object detection, it also provides valuable insights into how YOLOv4 detects objects.

In the digital marketplaces, businesses can micro-monitor sales worldwide and in real-time. Due to the vast amounts of data, there is a pressing need for tools that automatically highlight changing trends and anomalous (outlier) behavior that is potentially interesting to users. In collaboration with Danmark Music Group Ltd. we developed an unsupervised system for this problem based on a predictive neural network. To make the method transparent to developers and users (musicians, music managers, etc.), the system delivers two levels of outlier explanations: the deviation from the model prediction, and the explanation of the model prediction. We demonstrate both types of outlier explanations to provide value to data scientists and developers during development, tuning, and evaluation. The quantitative and qualitative evaluation shows that the users find the identified trends and anomalies interesting and worth further investigation. Consequently, the system was integrated into the production system. We discuss the challenges in unsupervised parameter tuning and show that the system could be further improved with personalization and integration of additional information, unrelated to the raw outlier score.