• 演讲人： 柳松（英国布里斯托大学 统计科学高级讲师）
• 时间：2022年09月14日 周三上午10:00
• 地点：浙江大学紫金港校区1416报告厅
• 主办单位：浙江大学数据科学研究中心

In this paper, we show that the arc length of the optimal ROC curve is an .-divergence By leveraging this result, we express the arc length using a variational objective and estimate it accurately using positive and negative samples. We show that this estimator has a non-parametric convergence rate ( depends on the smoothness). Using the same technique, we show that the area sandwiched between the optimal ROC curve and the diagonal can be expressed via a similar variational objective. These new insights lead to a novel two-step classification procedure that maximizes an approximate lower bound of the maximal AUC. Experiments on CIFAR-10 datasets show that the proposed two-step procedure achieves good AUC performance in imbalanced binary classification tasks while being less computationally demanding than the classic AUC maximizer in the offline setting.

报告人简介：

柳松，博士毕业于东京工业大学（导师杉山将）。毕业后曾在日本数理统计研究所担任博士后研究员（导师福水键次）。目前工作于英国布里斯托大学数学系，担任统计科学高级讲师。他的主要研究方向包括概率密度比估计，信息散度，统计的机器学习。发表论文等请参见: http://allmodelsarewrong.net.

联系人：孙文光（wgsun@zju.edu.cn）