arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2007.13982cs.LGstat.ML
收藏

潜在协变量混合模型的分布鲁棒损失

Distributionally Robust Losses for Latent Covariate Mixtures

Stanford University(斯坦福大学);Columbia Business School(哥伦比亚商学院)

John Duchi, Tatsunori Hashimoto, Hongseok Namkoong

摘要:

虽然现代大规模数据集通常由异质的子群体组成-例如,多个人口统计组或多个文本语料库-最小化平均损失的标准实践无法保证所有子群体的均匀低损失。我们提出了一个凸的过程,控制最坏情况下的性能在一个给定的大小的所有子群。我们的程序带有有限样本(非参数)收敛保证最差的子群体。从经验上讲,我们观察到词汇相似性,葡萄酒质量和累犯预测任务,我们的最坏情况程序学习模型,以及对看不见的亚群。

英文摘要:

While modern large-scale datasets often consist of heterogeneous subpopulations -- for example, multiple demographic groups or multiple text corpora -- the standard practice of minimizing average loss fails to guarantee uniformly low losses across all subpopulations. We propose a convex procedure that controls the worst-case performance over all subpopulations of a given size. Our procedure comes with finite-sample (nonparametric) convergence guarantees on the worst-off subpopulation. Empirically, we observe on lexical similarity, wine quality, and recidivism prediction tasks that our worst-case procedure learns models that do well against unseen subpopulations.

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。

↑