博客笔记:
1. 几篇learning with noisy labels论文的总结
博客地址:https://blog.csdn.net/ferriswym/article/details/89452430#t3
笔记:
相关论文阅读计划:
- Probabilistic End-to-end Noise Correction for Learning with Noisy Labels
链接:https://arxiv.org/pdf/1903.07788.pdf
状态:已读
概述:本文通过多次迭代修改Noisy data的标签分布来做到清洗数据的目的,不需要提前准备干净数据集或对noise的先验知识。借鉴了CVPR2018的Joint Optimization Framework for Learning with Noisy Labels这篇文章,两者都是用标签分布来代替标签,在模型训练时进行parameter learning和label learning。两篇的不同点在于进行label distribution更新时,CVPR2018那篇的更新策略较简单,而这篇的更新策略中用到了三种loss的组合,体现了end-to-end的思想。
状态:未读
- Generalized Cross Entropy Loss for Training DeepNeural Networks with Noisy Labels
状态:未读
- Understanding deep learning requires rethinking generalization
链接:https://arxiv.org/pdf/1611.03530.pdf
状态:未读
- Making Deep Neural Networks Robust to Label Noise: a Loss Correction Approach
链接:https://arxiv.org/pdf/1609.03683.pdf
状态:未读
- Training Convolutional Networks with Noisy Labels
链接:https://arxiv.org/pdf/1406.2080.pdf
状态:未读