arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2012.14348cs.LG
收藏

深度神经网络的广义分位数损失

Generalized Quantile Loss for Deep Neural Networks

Playtika AI Research(Playtika人工智能研究院)

Dvir Ben Or, Michael Kolomenkin, Gil Shabat

摘要:

本文介绍了一种简单的方法来向回归神经网络添加计数(或分位数)约束,这样,给定训练集中的$n$样本,它保证$m<n$样本的预测将大于实际值(标签)。与标准分位数回归网络不同,该方法可以应用于任何损失函数,而不一定是标准分位数回归损失,它使平均绝对差最小化。由于该计数约束几乎处处具有零梯度,因此无法使用标准梯度下降方法对其进行优化。为了克服这一问题,提出了一种基于标准神经网络优化过程的交替方案,并进行了理论分析。

英文摘要:

This note presents a simple way to add a count (or quantile) constraint to a regression neural net, such that given $n$ samples in the training set it guarantees that the prediction of $m<n$ samples will be larger than the actual value (the label). Unlike standard quantile regression networks, the presented method can be applied to any loss function and not necessarily to the standard quantile regression loss, which minimizes the mean absolute differences. Since this count constraint has zero gradients almost everywhere, it cannot be optimized using standard gradient descent methods. To overcome this problem, an alternation scheme, which is based on standard neural network optimization procedures, is presented with some theoretical analysis.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑