Linzh Zhao

New clipping method improves fairness in private machine learning

Researchers have developed a new method called bounded adaptive clipping to address disparate impacts in differentially private machine learning. Standard adaptive clipping can disproportionately suppress gradients from…