🏢 Department of Computer Science, Duke University

Noise in data surprisingly simplifies machine learning models, improving their interpretability without sacrificing accuracy; this paper quantifies this effect across various hypothesis spaces.