A technique to break down what a model learns internally into individual concepts or features it uses to make decisions.