知识卡片
跨数据样本价值不对称
将某来源样本折算成目标测试域的等效样本数,是任务和评估条件下的经验比较。书中 2011 年汽车分类表按目标市场列出不对称比例;零值不意味着该来源对所有任务没有信息。使用这类表必须带上年代、类别、模型和目标测试分布。
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
In table 35.2, each number represents the value of one training example from one dataset when used to test on another. A value of 0.25 means that you need 4 times more training examples than when using a sample derived from the same testing distribution. These results are from 2011 [471] and computer vision datasets have evolved since then.
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:摘录中关于跨数据样本价值不对称的定义、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:In table 35.2, each number represents the value of one training example from one dataset when used to test on another. A value of 0.25 means that you need 4 times more training examples than when using a sample derived from the same testing distribution. These results are from 2011 [471] and computer vision datasets have evolved since then.