An exploration of how training set composition bias in machine learning affects identifying rare objects

An exploration of how training set composition bias in machine learning affects identifying rare objects
复制标题

探索机器学习中的训练集组成偏差如何影响稀有物体的识别

DOI:
10.1016/j.ascom.2022.100617
复制
发表时间:
2022-07
影响因子:
2.5
通讯作者:
Tsai Chao-Wei
Tsai Chao-Wei
中科院分区:
物理与天体物理4区
文献类型:
--
作者:
Lake Sean E.;Tsai Chao-Wei

文献摘要

参考文献

相似文献

在训练机器学习分类器的数据时,其中一个类本质上是罕见的,分类器通常会为罕见的类分配太少的源。为了解决这个问题,通常会增加稀有类的示例权重,以确保它不会被忽略。由于同样的原因,在源类型的平衡更接近相等的受限数据上进行训练也是一种常见的做法。在这里,我们展示了这些实践可以使模型偏向于将资源过度分配给稀有类。我们还探讨了如何检测训练数据偏差何时对训练模型的预测产生统计上显著的影响,以及如何减少偏差的影响。虽然这里所开发的技术的影响程度会随着应用程序的细节而变化,但在大多数情况下应该是适度的。然而,它们在每次使用机器学习分类模型时都是普遍适用的,这使得它们类似于贝塞尔对样本方差的校正。
When training a machine learning classifier on data where one of the classes is intrinsically rare, the classifier will often assign too few sources to the rare class. To address this, it is common to up-weight the examples of the rare class to ensure it isn't ignored. It is also a frequent practice to train on restricted data where the balance of source types is closer to equal for the same reason. Here we show that these practices can bias the model toward over-assigning sources to the rare class. We also explore how to detect when training data bias has had a statistically significant impact on the trained model's predictions, and how to reduce the bias's impact. While the magnitude of the impact of the techniques developed here will vary with the details of the application, for most cases it should be modest. They are, however, universally applicable to every time a machine learning classification model is used, making them analogous to Bessel's correction to the sample variance.
DOI: 10.1111/j.1365-2966.2007.12040.x
发表时间: 2006-04
影响因子: 4.8
作者:
A. Lawrence;S. Warren;O. Almaini;A. Edge;N. Hambly;R. Jameson;P. Lucas;M. Casali;A. Adamson
通讯作者: A. Lawrence;S. Warren;O. Almaini;A. Edge;N. Hambly;R. Jameson;P. Lucas;M. Casali;A. Adamson
DOI: 10.1093/mnras/stz2947
发表时间: 2019-10
影响因子: 4.8
作者:
C. Bailer-Jones;M. Fouesneau;R. Andrae
通讯作者: C. Bailer-Jones;M. Fouesneau;R. Andrae
DOI: 10.1088/0004-6256/143/1/7
发表时间: 2011-11
期刊: The Astronomical Journal
影响因子: --
作者:
S. Lake;E. L. Wright;S. Petty;R. Assef;T. Jarrett;S. Stanford;D. Stern;C. Tsai
通讯作者: S. Lake;E. L. Wright;S. Petty;R. Assef;T. Jarrett;S. Stanford;D. Stern;C. Tsai
DOI: 10.1007/978-3-662-44185-5_5314
发表时间: 2021-12
期刊: Encyclopedia of Astrobiology
影响因子: --
作者:
J. Wagg;I. Jiménez-Serra;Tyler Bourke;Robert Braun;Phil Diamond;William Garnier;Jimi Green;
通讯作者: J. Wagg;I. Jiménez-Serra;Tyler Bourke;Robert Braun;Phil Diamond;William Garnier;Jimi Green;
DOI: 10.1093/mnras/stv1242
发表时间: 2015-08-21
影响因子: 4.8
作者:
Petroff, E.;Keane, E. F.;Bhandari, S.
通讯作者: Bhandari, S.