知识卡片

普通读书笔记卡 · 6713.a · 生成模型:GPT-6

卷积共享带来平移等变性

理解等级:中级

相同局部模式移到另一位置时,共享卷积核可在相应位置产生相似响应,通常称平移等变。结合全局池化可减少对响应位置的依赖,但边界、步长和填充会影响结果;普通卷积不天然实现任意词距伸缩的尺度不变。

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 main advantage of CNNs over older NLP algorithms is that they can recognize patterns no matter where they occur in the text (translation invariance) as well as how spread out they are (scale invariance). Term frequency–inverse document frequency (TF–IDF) vectors don’t have any way of recognizing and generalizing from patterns in your text, and fully connected neural networks overgeneralize from particular patterns at particular

    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:摘录中关于卷积共享带来平移等变性的定义、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 main advantage of CNNs over older NLP algorithms is that they can recognize patterns no matter where they occur in the text (translation invariance) as well as how spread out they are (scale invariance). Term frequency–inverse document frequency (TF–IDF) vectors don’t have any way of recognizing and generalizing from patterns in your text, and fully connected neural networks overgeneralize from particular patterns at particular