知识卡片

普通读书笔记卡 · 9012.e · 生成模型:GPT-6

以组件清单比较迁移方法

理解等级:中级

比较迁移方法可逐项列出预训练组件、适配时更新变量、使用的数据及推理路径。微调更新主参数,探针更新读出,提示更新输入信号,蒸馏借教师输出,合成数据借生成分布。这些选择可以组合,名称不是相互排斥的完整系统规格。

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

      In table 37.1, we present the relationship these methods by framing them all as empirical risk minimization with different components optimized during pretraining and/or during adaptation. Table 37.1 Different kinds of transfer learning presented as empirical risk minimization (ERM). Notes: The ⊕ symbol indicates concatenation or addition (or other kinds of signal combination). Domain adaptation does not map neatly to ERM so we omit it

    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:摘录中关于以组件清单比较迁移方法的定义、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:In table 37.1, we present the relationship these methods by framing them all as empirical risk minimization with different components optimized during pretraining and/or during adaptation. Table 37.1 Different kinds of transfer learning presented as empirical risk minimization (ERM). Notes: The ⊕ symbol indicates concatenation or addition (or other kinds of signal combination). Domain adaptation does not map neatly to ERM so we omit it