Deep learning III - II Machine Learning Strategy 2 - Bias and Variance with mismatched data distribu

Posted dqhl1990

tags:

篇首语：本文由小常识网(cha138.com)小编为大家整理，主要介绍了Deep learning III - II Machine Learning Strategy 2 - Bias and Variance with mismatched data distribu相关的知识，希望对你有一定的参考价值。

Bias and Variance with mismatched data distribution 不相称数据分布的偏差和方差

  • 应对在mismatched data distribution情况下的模型评估，构建了development-train set作为中间评估手段。各个数据集之间的error反应的模型的问题如下图所示


  • 举个例子：

  • 如何消除由于data mismatch 造成的模型问题

    1. 先看看 dev-train set error 和 dev set error有差距的原因是什么
    2. 人工合成一些更靠近dev set的数据添加到train set中，但是要小心在人工合成中因为合成材料数量小而可能造成的模型overfitting到局部的情况。

以上是关于Deep learning III - II Machine Learning Strategy 2 - Bias and Variance with mismatched data distribu的主要内容，如果未能解决你的问题，请参考以下文章

Deep learning III - II Machine Learning Strategy 2 - Multi-task Learning 多任务学习

Deep learning III - II Machine Learning Strategy 2 - Mismatched training and dev/test set

Deep learning III - II Machine Learning Strategy 2 - Bias and Variance with mismatched data distribu

Deep learning II - III Batch Normalization - Normalizing activation in a network 激励函数输出归一化

Deep learning III - I Machine Learning Strategy - improving your model performance 如何减少可变避免的变差和方差

Deep learning III - I Machine Learning Strategy - train development and test sets 训练集开发集和测试集