知识卡片

普通读书笔记卡 · 9012.a · 生成模型:GPT-6

无监督域适配不使用目标标签

理解等级:中级

无监督域适配通常拥有源域标签与目标域无标签输入,用后者估计或对齐域差异。目标域样本可参与适配这一设定必须公开,并与最终独立评估的使用规则分清。没有目标标签不等于没有任何目标域信息,也不等于无需验证标签保持。

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

    • · 查看原书依据

      Commonly, we only have access to labels at training time, but may have plentiful unlabeled inputs at testing time. This setting is sometimes referred to as unsupervised domain adaptation. Here we cannot make use of test labels but we can still align the test inputs to look like the training inputs (or vice versa). One way to do so is to use a generative adversarial

    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:摘录中关于无监督域适配不使用目标标签的定义、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:Commonly, we only have access to labels at training time, but may have plentiful unlabeled inputs at testing time. This setting is sometimes referred to as unsupervised domain adaptation. Here we cannot make use of test labels but we can still align the test inputs to look like the training inputs (or vice versa). One way to do so is to use a generative adversarial