BriefGPT.xyz
Feb, 2019
通过正则化评估注释者混淆来从嘈杂的标签中学习
Learning From Noisy Labels By Regularized Estimation Of Annotator Confusion
HTML
PDF
Ryutaro Tanno, Ardavan Saeedi, Swami Sankaranarayanan, Daniel C. Alexander, Nathan Silberman
TL;DR
本文研究了如何解决多个注释器提供的有噪声的标签的问题,提出了一种同时学习个体注释器模型和真实标签分布的方法,并通过正则化项促进收敛于真实注释器混淆矩阵的方法,在图像分类任务上实验结果表明,该方法能够估计注释器的技能并表现出良好的性能。
Abstract
The predictive performance of
supervised learning algorithms
depends on the quality of labels. In a typical
label collection process
, multiple annotators provide subjective noisy estimates of the "truth" under th
→