利用梯度衍生的度量对不同 ially private 训练中的数据选择和估值进行优化（英文版）

英文标题：Leveraging gradient-derived metrics for data selection and valuation in differentially private training中文摘要：研究了如何在严格保护隐私的情况下，利用梯度信息来选择有利于模型训练的数据，解决在协同训练深度学习模型中，难以区分出有用数据点的问题。英文摘要：Obtaining high-quality data for collaborative training of machine learningmodels can be a challenging task due to A) the regulatory concerns and B) lackof incentive to participate. The first issue can be addressed through the useof privacy enhancing technologies (PET), one of the most frequently used onebeing diff