Hotspot analysis highlights areas of an image or features in data that are important. It is often used in explainable AI applied to machine learning models such as a neural network; pixels or features are highlighted that are particularly critical in determining the output. These hotspots can sometimes be derived by modifying the algorithm used in the model, but perturbation methods can be used with a black-box model.
Used in Chap. 21: pages 331, 337
Used in glossary entries: black-box machine learning, explainable AI, machine learning, neural network, perturbation techniques
Links:
alandix.com: XAI Kitbag
Key feature detection through perturbations