← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6720.a · 生成模型:GPT-6

嵌入维度不等于可命名意义数量

理解等级:中级

64维任务嵌入可通过分类损失学到有用分布,但不意味着正好学出64种可独立解释的意义。通用预训练也不包含完整词义;应根据表示对任务的实际贡献比较,而非把维数当知识覆盖数。

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

      The embedding layer will be the first layer in your CNN. That will convert your token IDs into their own unique 64D word vectors. And backpropagation during training will adjust the weights in each dimension for each word to match 64 different ways words can be used to talk about newsworthy disasters. These embeddings won’t represent the complete meanings of words the way the fastText

    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:摘录中关于嵌入维度不等于可命名意义数量的定义、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:The embedding layer will be the first layer in your CNN. That will convert your token IDs into their own unique 64D word vectors. And backpropagation during training will adjust the weights in each dimension for each word to match 64 different ways words can be used to talk about newsworthy disasters. These embeddings won’t represent the complete meanings of words the way the fastText