← 返回起始笔记位置

知识卡片

普通读书笔记卡 · 8859.a · 生成模型:GPT-6

多组最优参数暴露隐式选择

理解等级:高级

数据与目标不能唯一确定参数时,多个候选可获得相同训练损失。最终选中哪个解还受初始化、优化器和数值实现影响。高阶多项式示例同时涉及欠定和条件数问题,不能在秩亏时仍把普通逆公式视为唯一且稳定的优化器。

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

      there is insufficient data for the objective to uniquely identify one of the hypotheses to be the best. Because of this, the hypothesis output by the optimizer may be an arbitrary one, or rather will be due to details of the optimization algorithm (e.g., how it is initialized), rather than selected by the objective. The optimizer we used above has a tendency to pick, among

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第11章《泛化问题》(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/c011.xhtml) - 结论依据:摘录中关于多组最优参数暴露隐式选择的定义、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:there is insufficient data for the objective to uniquely identify one of the hypotheses to be the best. Because of this, the hypothesis output by the optimizer may be an arbitrary one, or rather will be due to details of the optimization algorithm (e.g., how it is initialized), rather than selected by the objective. The optimizer we used above has a tendency to pick, among