知识卡片
外部生成器带来先验
用外部大数据预训练的生成器增强小型目标集,会把外部学习到的结构引入任务,区别于只在同一小数据上自举生成。收益仍取决于生成器与目标域的匹配、合成标签正确性及重复偏差;记录生成模型来源,并用真实目标样本评估。
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
A clearer gain can be achieved by using a gθ pretrained on external data. For example, if we have a general purpose image generative model, trained on millions of images (such as [397] or [409]), then we can use it to augment a small dataset for a new specific task we come across.
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第36章《鲁棒性与通用性训练》(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/c036.xhtml)
- 结论依据:摘录中关于外部生成器带来先验的定义、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:A clearer gain can be achieved by using a gθ pretrained on external data. For example, if we have a general purpose image generative model, trained on millions of images (such as [397] or [409]), then we can use it to augment a small dataset for a new specific task we come across.