Outlier-Aware Training for Improving Group Accuracy Disparities
Outlier-Aware Training for Improving Group Accuracy Disparities
复制标题
DOI：
10.48550/arxiv.2210.15183
复制
发表时间：
2022-10
期刊：
影响因子：
--
通讯作者：
Li-Kuang Chen;Canasai Kruengkrai;J. Yamagishi
中科院分区：
文献类型：
--
作者：
Li-Kuang Chen;Canasai Kruengkrai;J. Yamagishi
处理虚假相关性的方法，如Just Train Twice （JTT, Liu et al. 2021）涉及重新加权训练集的一个子集，以最大化最差组的准确性。然而，重新加权的示例集可能包含阻碍模型学习的不可学习示例。我们建议通过检测训练集的异常值并在重新加权之前删除它们来缓解这种情况。实验结果表明，该方法具有与JTT相当甚至更好的精度，可以检测和去除JTT中被重加权子集中的标注错误。
Methods addressing spurious correlations such as Just Train Twice (JTT, Liu et al. 2021) involve reweighting a subset of the training set to maximize the worst-group accuracy. However, the reweighted set of examples may potentially contain unlearnable examples that hamper the model’s learning. We propose mitigating this by detecting outliers to the training set and removing them before reweighting. Our experiments show that our method achieves competitive or better accuracy compared with JTT and can detect and remove annotation errors in the subset being reweighted in JTT.