知识卡片
域适配需保持任务语义
可把目标域输入转换为源域风格,再使用源预测器;也可先把源数据转换向目标域再训练。仅对齐输入边缘分布 p(x) 不足以保证预测正确:映射可能改变类别,或两域 p(y|x) 不同。应同时约束任务语义并检查目标域表现。
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
- · 迁移学习与适配
原书出处 · 1 处
- · 查看原书依据
The first option is to adapt the test data, to look more like the training data. Then f should work well on both the training data and the test data. Or we can go the other way around, adapting the training data to look like the test data, before training f on the adapted training data. In either case, the trick is to make the
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第37章《迁移学习与适配》(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/c037.xhtml)
- 结论依据:摘录中关于域适配需保持任务语义的定义、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:The first option is to adapt the test data, to look more like the training data. Then f should work well on both the training data and the test data. Or we can go the other way around, adapting the training data to look like the test data, before training f on the adapted training data. In either case, the trick is to make the