知识卡片

普通读书笔记卡 · 8981.a · 生成模型:GPT-6

记住训练集不等于泛化

理解等级:中级

从训练样本记忆库随机抽取可生成真实外观,却可能只是拟合有限经验分布。若目标是新样本,需要检验模型在未见数据和变化上的表现,以及是否反复复制。函数简单或平滑只是可能的归纳偏好,不能单独保证不记忆或正确泛化。

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

      What if our model just memorizes all the training examples and generates random draws from this memory? Indeed, memorized training samples perfectly satisfy the goal of making synthetic data that looks real. A second, sometimes overlooked, property of a good generative model is that it be a simple, or smooth, function, so that it generalizes to producing synthetic samples that look like the training data

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第32章《生成模型》(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/c032.xhtml) - 结论依据:摘录中关于记住训练集不等于泛化的定义、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:What if our model just memorizes all the training examples and generates random draws from this memory? Indeed, memorized training samples perfectly satisfy the goal of making synthetic data that looks real. A second, sometimes overlooked, property of a good generative model is that it be a simple, or smooth, function, so that it generalizes to producing synthetic samples that look like the training data