17 Sparse and Faithful Explanations Without Sparse Models

Oregon Convention Center 

Even if a model is not globally sparse, it is possible for decisions made from that model to be accurately and faithfully described by a small number of features. We introduce the Sparse Explanation Value (SEV), a new way of measuring sparsity in machine learning models. SEV is a measure of decision sparsity rather than overall model sparsity. We introduce algorithms that reduce SEV without sacrificing accuracy, providing sparse and completely faithful explanations, even without globally sparse models.