知识卡片

普通读书笔记卡 · 6724.a · 生成模型:GPT-6

线性输出层按目标决定宽度和损失

理解等级:中级

拼接特征经线性层产生目标所需输出,二元可用一个logit,多类通常用每类logit。再按损失契约选择sigmoid、softmax或阈值;argmax会给类别编号而非保留同宽向量,不能把各输出步骤都写成相同shape。

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

      7.5.3 Linear layers The concatenated encodings approach gave you a lot of information about each microblog post—the encoding vector had 1,856 values, and the largest word vectors you worked with in chapter 6 had 300 dimensions—but all you really want to learn from this pipeline is the binary answer to our overarching question: Is it newsworthy or not? Do you remember when you were trying

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第07章《卷积网络识别文本模式》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-7.html) - 结论依据:摘录中关于线性输出层按目标决定宽度和损失的定义、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:7.5.3 Linear layers The concatenated encodings approach gave you a lot of information about each microblog post—the encoding vector had 1,856 values, and the largest word vectors you worked with in chapter 6 had 300 dimensions—but all you really want to learn from this pipeline is the binary answer to our overarching question: Is it newsworthy or not? Do you remember when you were trying