2020/09/21 by Teodora Popordanoska, Mohit Kumar, Popordanoska, Teodora +4 · 2 citations
Computer Science · #Artificial Intelligence (cs.AI) #Data Stream Mining Techniques #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Data Classification
paper · pdf · doi:10.48550/arxiv.2009.09723
openalex publication_date 2020/09/21 · openalex created_date 2022/10/05 · openalex updated_date 2026/07/28
We introduce explanatory guided learning (XGL), a novel interactive learning\nstrategy in which a machine guides a human supervisor toward selecting\ninformative examples for a classifier. The guidance is provided by means of\nglobal explanations, which summarize the classifier's behavior on different\nregions of the instance space and expose its flaws. Compared to other\nexplanatory interactive learning strategies, which are machine-initiated and\nrely on local explanations, XGL is designed to be robust against cases in which\nthe explanations supplied by the machine oversell the classifier's quality.\nMoreover, XGL leverages global explanations to open up the black-box of\nhuman-initiated interaction, enabling supervisors to select informative\nexamples that challenge the learned model. By drawing a link to interactive\nmachine teaching, we show theoretically that global explanations are a viable\napproach for guiding supervisors. Our simulations show that explanatory guided\nlearning avoids overselling the model's quality and performs comparably or\nbetter than machine- and human-initiated interactive learning strategies in\nterms of model quality.\n