知识卡片
用交叉域误差矩阵
对来源 A、B 分别训练,再分别在 A、B 测试,形成四格误差矩阵,区分域内拟合与跨域泛化。本书五样本回归案例的 A→A/A→B/B→A/B→B 误差约为 0.18⁄0.70⁄1.93⁄0.23;它展示迁移方向不对称,不是可复用的性能指标。
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
When training and testing is done using samples from the same distribution (A to A, or B to B) the learned model (best fit) passes near the test samples and the error is small. However, when training and test are coming from different distributions, the error can be large. Table 35.1 shows the generalization error made with different combinations of training and set sets as
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:摘录中关于用交叉域误差矩阵的定义、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:When training and testing is done using samples from the same distribution (A to A, or B to B) the learned model (best fit) passes near the test samples and the error is small. However, when training and test are coming from different distributions, the error can be large. Table 35.1 shows the generalization error made with different combinations of training and set sets as