← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9113.a · 生成模型:GPT-6

下游新场景未必预训练域外

理解等级:中级

某图不属于手工简单世界,不代表也超出大型模型的训练分布;自然图像可能已被广泛覆盖。声称域外泛化时应说明相对于哪个训练集或假设集合,并考虑数据重合,不能只依据测试者感觉陌生。

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

      Figure 55.5: Out of domain generalization. For the pretrained model, these images might still be within the domain of the training, as it contained a very diverse set of natural images. The simple system model only works for the first example and breaks down for the rest, producing completely wrong reconstructions.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第55章《重访简单视觉系统》(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/c055b.xhtml) - 结论依据:摘录中关于下游新场景未必预训练域外的定义、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:Figure 55.5: Out of domain generalization. For the pretrained model, these images might still be within the domain of the training, as it contained a very diverse set of natural images. The simple system model only works for the first example and breaks down for the rest, producing completely wrong reconstructions.