知识卡片
样本价值依赖测试域
相同数量的训练样本,对特定测试分布的贡献可能不同。本书共享底层二次函数、输入高斯均值和尺度不同的玩具实验中,B 域约需十倍样本达到 A 域训练的某种表现。这个比例仅属于该设置;样本价值依赖目标域、模型和性能区间,不能给数据来源贴固定价格。
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
Interestingly, both training sets (A and B) succeed in reducing the generalization error on the test set of dataset A when more training data is available. In this example, even the biased dataset B is still able to train a successful model of dataset A. However, the reduction of the error is slower than when training on dataset A. Examples from dataset B are less
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:摘录中关于样本价值依赖测试域的定义、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:Interestingly, both training sets (A and B) succeed in reducing the generalization error on the test set of dataset A when more training data is available. In this example, even the biased dataset B is still able to train a successful model of dataset A. However, the reduction of the error is slower than when training on dataset A. Examples from dataset B are less