← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 7006.k · 生成模型:DeepSeek-V4.1-Flash

标签、人工标注与数据增强

理解等级:中级

标签是监督学习的另一类训练数据:特征是模型输入,标签则是正确输出的示例,训练时用于设定模型参数,留出的标签还用于评估质量。推荐系统这类问题的标签可由日志关联订单算法化生成,比人工标注更一致准确,也常与特征一起存进特征存储。语音转写、图像分类、判断钩针针法这类任务必须靠人,成本很高:要培训标注员,同一张图常需多人标注才可信。为放大昂贵标签的价值,常用数据增强(fuzzing)做缩放、裁剪、加噪等不改变标签正确性的变换来扩充样本;但增强应由训练系统完成,且切莫把同源图分别放进训练集与测试集。

这张卡片看不懂?

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

    • Feature and Training Data · 查看原书依据

      Because of the large cost associated with acquiring human-generated labels, training systems should be designed to get as much mileage from them as possible. One technique commonly used is data augmentation: the feature data is “fuzzed” in a way that changes the features but doesn’t change the correctness of the label.5 For example, consider our stitch classification problem.

    参考来源

    - 位置:《Reliable Machine Learning》第04章《Feature and Training Data》(源文件:extracted/Reliable Machine Learning/_epub-src/OEBPS/ch04.xhtml) - 结论依据:原文明确阐述该主题,并给出定义、实践或反例。 - 原始内容:Because of the large cost associated with acquiring human-generated labels, training systems should be designed to get as much mileage from them as possible. One technique commonly used is data augmentation: the feature data is “fuzzed” in a way that changes the features but doesn’t change the correctness of the label.5 For example, consider our stitch classification problem.