知识卡片
学习曲线检验扩数效果
跨域误差高时,比较固定目标测试集上随训练样本数变化的曲线,判断增加现有来源数据是否仍有效,以及和目标域数据差多少。曲线是否下降或饱和取决于分布支持、假设空间等条件;本书两高斯共享函数的结果不能推出分布不重叠时扩数也能补救。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
原书出处 · 1 处
- · 查看原书依据
It is interesting in general to look at the performance as a function of training data. If we add more data, does performance improve when training and testing on different datasets? Or does it saturate? The answer will depend on the form of the data distributions and the hypothesis space. But we can check what happens in this toy example.
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第35章《数据偏差与分布漂移》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c035b.xhtml)
- 结论依据:摘录中关于学习曲线检验扩数效果的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:It is interesting in general to look at the performance as a function of training data. If we add more data, does performance improve when training and testing on different datasets? Or does it saturate? The answer will depend on the form of the data distributions and the hypothesis space. But we can check what happens in this toy example.