Learning High Coverage Discriminative Parsimonious Rulesets

The paper introduces CDPR, a novel framework for learning high-coverage, discriminative, and parsimonious rule sets for classification tasks. It presents two algorithms based on submodular maximization that ensure provable guarantees on coverage while enhancing both accuracy and interpretability of the learned rules. This advancement is significant for practitioners as it addresses the trade-off between predictive accuracy and interpretability, offering a more effective solution for deploying interpretable AI models.