知识卡片
损失与优化器也能迁移
迁移对象可以超出模型权重:预训练损失函数或优化器,再用于新任务,也是在复用学习得到的结构。描述时应明确学的是哪个组件、预训练目标、适配时是否冻结以及推理是否保留它;同一经验风险形式可能隐藏不同知识来源。
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
This chapter has covered a few different kinds of knowledge that can be pretrained and transferred. We have not aimed to be exhaustive. As an exercise, you can try the following: (1) download the code for your favorite learning algorithm, (2) pick a random line, (3) and figure out how you can pretrain whatever is being computed on that line. Maybe you picked the line
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:摘录中关于损失与优化器也能迁移的定义、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:This chapter has covered a few different kinds of knowledge that can be pretrained and transferred. We have not aimed to be exhaustive. As an exercise, you can try the following: (1) download the code for your favorite learning algorithm, (2) pick a random line, (3) and figure out how you can pretrain whatever is being computed on that line. Maybe you picked the line