Principles of Explanatory Debugging to Personalize Interactive Machine Learning

摘要: How can end users efficiently influence the predictions that machine learning systems make on their behalf? This paper presents Explanatory Debugging, an approach in which system explains to how it made each of its predictions, and user then any necessary corrections back system. We present principles underlying this a prototype instantiating it. An empirical evaluation shows Debugging increased participants' understanding by 52% allowed participants correct mistakes up twice as using traditional